2023 data, ML and AI landscape: ChatGPT, generative AI and more

Noteworthy financings in generative AI

OpenAI received a $10B investment from Microsoft in January 2023; Runway ML, an AI-powered video editing platform, raised a $50M Series C at a $500M valuation in December 2022; ImagenAI, an AI-powered photo editing and post-production automation startup, raised $30 million in December 2022; Descript, and AI-powered media editing app, raised $50M in its Series C in November 2022; Mem, an AI-powered note-taking app, raised $23.5M in its Series A in November 2022; Jasper AI, an AI-powered copywriter, raised $125M at a $1.5B valuation in October 2022; Stability AI, the generative AI company behind Stable Diffusion, raised $101M at $1B valuation in October 2022; You, an AI-powered search engine, raised $25M in its Series A financings; Hugging Face, a repository of open source machine learning models, raised $100M in its Series C at a $1B valuation in May 2022; Inflection AI, AGI startup, raised $225M in its first round of equity financing in May 2022; Anthropic, an AI research firm, raised $580M in its Series B (investors including from SBF and Caroline Ellison!) in April 2022; Cohere, an NLP platform, raised $125M in its Series B in February 2022.

Expect a lot more of this. Cohere is reportedly in talks to raise hundreds of millions of dollars in a funding round that could value the startup at more than $6 billion

M&A

2022 was a difficult year for acquisitions, punctuated by the failed $40B acquisition of ARM by Nvidia (which would have affected the competitive landscape of everything from mobile to AI in data centers). The drawdown in the public markets, especially tech stocks, made acquisitions with any stock component more expensive compared to 2021. Late-stage startups with strong balance sheets, on the other hand, generally favored reducing burn instead of making splashy acquisitions. Overall, startup exit values fell by over 90% year over year to $71.4B from $753.2B in 2021.

That said, there were several large acquisitions and a number of (presumably) small tuck-in acquisitions, a harbinger of things to come in 2023, as we expect many more of those in the year ahead (we discuss consolidation in Part III on Data Infrastructure).

Private equity firms may play an outsized role in this new environment, whether on the buy or sell side.   Qlik just announced its intent to acquire Talend. This is notable because both companies are owned by Thoma Bravo, who presumably played marriage broker. Progress also just completed its acquisition of MarkLogic, a NoSQL database provider MarkLogic for $355M.  MarkLogic, rumored to have revenues “around $100M”, was owned by private equity firm Vector Capital Management.

MAD 2023, Part III: Data infrastructure back to reality

In the hyper-frothy environment of 2019-2021, the world of data infrastructure (nee Big Data) was one of the hottest areas for both founders and VCs.

It was dizzying and fun at the same time, and perhaps a little weird to see so much market enthusiasm for products and companies that are ultimately very technical in nature.

Regardless, as the market has cooled down, that moment is over. While good companies will continue to be created in any market cycle, and “hot” market segments will continue to pop up, the bar has certainly escalated dramatically in terms of differentiation and quality for any new data infrastructure startup to get real interest from potential customers and investors.

Here is our take on some of the key trends in the data infra market in 2023. The first couple trends are higher level and should be interesting to everyone, the others are more in the weeds:

• Brace for impact: bundling and consolidation 
• The Modern Data Stack under pressure 
• The end of ETL?
• Reverse ETL vs CDP
• Data mesh, products, contracts: dealing with organizational complexity
• [Convergence]
• Bonus: What impact will AI have on data and analytics?

Brace for impact: Bundling and consolidation 

If there’s one thing the MAD landscape makes obvious year after year, it’s that the data/AI market is incredibly crowded. In recent years, the data infrastructure market was very much in “let a thousand flowers bloom” mode.  

The Snowflake IPO (the biggest software IPO ever) acted as a catalyst for this entire ecosystem. Founders started literally hundreds of companies, and VCs happily funded them (again, and again, and again) within a few months. New categories (e.g., reverse ETL, metrics stores, data observability) appeared and became immediately crowded with a number of hopefuls.

On the customer side, discerning buyers of technology, often found in scale-ups or public tech companies, were willing to experiment and try the new thing with little oversight from the CFO office. This resulted in many tools being tried and purchased in parallel. 

Now, the music has stopped. 

On the customer side, buyers of technology are under increasing budget pressure and CFO control. While data/AI will remain a priority for many, even during a recessionary period, they have too many tools as it is, and they’re being asked to do more with less.  They also have fewer resources to engineer anything. They’re less likely to be experimental or work with immature tools and unproven startups. They’re more likely to pick established vendors that offer tightly integrated suites of products, stuff that “just works.”

This leaves the market with too many data infrastructure companies doing too many overlapping things.

In particular, there’s an ocean of “single-feature” data infrastructure (or MLOps) startups (perhaps too harsh a term, as they’re just at an early stage) that are going to struggle to meet this new bar.  Those companies are typically young (1-4 years in existence), and due to limited time on earth, their product is still largely a single feature, although every company hopes to grow into a platform; they have some good customers but not a resounding product-market-fit just yet.

This class of companies has an uphill battle in front of them and a tremendous amount of growing to do in a context where buyers are going to be weary and VC cash is scarce.

Expect the beginning of a Darwinian period ahead. The best (or luckiest, or best funded) of those companies will find a way to grow, expand from a single feature to a platform (say, from data quality to a full data observability platform), and deepen their customer relationships. 

Others will be part of an inevitable wave of consolidation, either as a tuck-in acquisition for a bigger platform or as a startup-on-startup private combination. Those transactions will be small, and none of them will produce the kind of returns founders and investors were hoping for. (we are not ruling out the possibility of multi-billion dollar mega deals in the next 12-18 months, but those will most likely require the acquirers to see the light at the end of the tunnel in terms of the recessionary market). 

Still, consolidation will be better than simply going out of business. Bankruptcy, an inevitable part of the startup world, will be much more common than in the last few years, as companies cannot raise their next round or find a home. 

At the top of the market, the larger players have already been in full product expansion mode. It’s been the cloud hyperscaler’s strategy all along to keep adding products to their platform. Now Snowflake and Databricks, the rivals in a titanic shock to become the default platform for all things data and AI (see the 2021 MAD landscape), are doing the same.

Databricks seems to be on a mission to release a product in just about every box of the MAD landscape. This product expansion has been done almost entirely organically, with a very small number of tuck-in acquisitions along the way – Datajoy and Cortex Labs in 2022. Snowflake has also been releasing features at a rapid pace. It has become more acquisitive as well. It announced three acquisitions in the first couple of months of 2023 already.

Confluent, the public company built on top of the open-source streaming project Kafka, is also making interesting moves by expanding to Flink, a very popular streaming processing engine. It just acquired Immerok. This was a quick acquisition, as Immerok was founded in May 2022 by a team of Flink committees and PMC members, funded with $17M in October and acquired in January 2023. 

Some slightly smaller but still unicorn-type startups are also starting to expand aggressively, starting to encroach on other’s territories in an attempt to grow into a broader platform.

As an example, transformation leader dbt Labs first announced a product expansion into the adjacent semantic layer area in October 2022. Then, it acquired an emerging player in the space, Transform (dbt’s blog post provides a nice overview of the semantic layer and metrics store concept) in February 2023.

Some categories in data infrastructure feel particularly ripe for consolidation of some sort – the MAD landscape provides a good visual aid for this, as the potential for consolidation maps pretty closely with the fullest boxes:

ETL and reverse ETL: Over the last three or four years, the market has funded a good number of ETL startups (to move data into the warehouse), as well as a separate group of reverse ETL startups (to move data out of the warehouse).  It is unclear how many startups the market can sustain in either category. Reverse ETL companies are under pressure from different angles (see below), and it is possible that both categories may end up merging.  ETL company Airbyte acquired Reverse ETL startup Grouparoo. Several companies like Hevo Data position as end-to-end pipelines, delivering both ETL and reverse ETL (with some transformation too), as does data syncing specialist Segment. Could ETL market leader FIvetran acquire or (less likely) merge with one of its Reverse ETL partners like Census or Hightouch?

Data quality and observability: The market has seen a glut of companies that all want to be the “Datadog of data.” What Datadog does for software (ensure reliability and minimize application downtime), those companies want to do for data – detect, analyze and fix all issues with respect to data pipelines. Those companies come at the problem from different angles: Some do data quality (declaratively or through machine learning), others do data lineage, and others do data reliability. Data orchestration companies also play in the space. Many of those companies have excellent founders, are backed by premier VCs and have built quality products. However, they are all converging in the same direction in a context where demand for data observability is still comparatively nascent.

Data catalogs: As data becomes more complex and widespread within the enterprise, there is a need for an organized inventory of all data assets.  Enter data catalogs, which ideally also provide search, discovery and data management capabilities. While there is a clear need for the functionality, there are also many players in the category, with smart founders and strong VC backing, and here as well, it is unclear how many the market can sustain. It is also unclear whether data catalogs can be separate entities outside of broader data governance platforms long term.

MLOps: While MLOps sits in the ML/AI section of the MAD landscape, it is also infrastructure and it is likely to experience some of the same circumstances as the above.  Like the other categories, MLOps plays an essential role in the overall stack, and it is propelled by the rising importance of ML/AI in the enterprise.  However, there is a large number of companies in the category, most of which are well-funded but early on the revenue front.  They started from different places (model building, feature stores, deployment, transparency, etc.), but as they try to go from single feature to a broader platform, they are on a collision course with each other. Also, many of the current MLOps companies have primarily focused on selling to scale-ups and tech companies.  As they go upmarket, they may start bumping into the enterprise AI platforms that have been selling to Global 2000 companies for a while, like Dataiku, Datarobot, H2O, as well as the cloud hyperscalers.

The modern data stack under pressure

A hallmark of the last few years has been the rise of the “Modern Data Stack” (MDS). Part architecture, part de facto marketing alliance amongst vendors, the MDS is a series of modern, cloud-based tools to collect, store, transform and analyze data. At the center of it, there’s the cloud data warehouse (Snowflake, etc.). Before the data warehouse, there are various tools (Fivetran, Matillion, Airbyte, Meltano, etc.) to extract data from their original sources and dump it into the data warehouse. At the warehouse level, there are other tools to transform data, the “T” in what used to be known as ETL (extract transform load) and has been reversed to ELT (here, dbt Labs reigns largely supreme). After the data warehouse, there are other tools to analyze the data (that’s the world of BI, for business intelligence) or extract the transformed data and plug it back into SaaS applications (a process known as “reverse ETL”).

Up until recently, the MDS was a fun but little world. As Snowflake’s fortunes kept rising, so did the entire ecosystem around it. Now, the world has changed.  As cost control becomes paramount, some may question the approach that is at the heart of the modern data stack: Dump all your data somewhere (a data lake, lakehouse or warehouse) and figure out what to do with it later, which turns out to be expensive and not always that useful. 

Now the MDS is under pressure. In a world of cost control and rationalization, it’s almost too obvious a target. It’s complex (as customers need to stitch everything together and deal with multiple vendors). It’s expensive (as every vendor wants their margin and also because you need an in-house team of data engineers to make it all work). And it’s arguably elitist (as those are the most bleeding-edge, best-in-breed tools, requiring customers to be sophisticated both technically and in terms of use cases), serving the needs of the few.

What happens when MDS companies stop being friendly and start competing with one another for smaller customer budgets?

As an aside, the complexity of the MDS has given rise to a new category of vendors that “package” various products under one fully managed platform (as mentioned above, a new box in the 2023 MAD featuring companies like Y42 or Mozart Data).  The underlying vendors are some of the usual suspects in MDS, but most of those platforms abstract away both the business complexity of managing several vendors and the technical complexity of stitching together the various solutions.  

The end of ETL?

As a twist on the above, there’s a parallel discussion in data circles as to whether ETL should even be part of data infrastructure going forward. ETL, even with modern tools, is a painful, expensive and time-consuming part of data engineering. 

At its Re:Invent conference last November, Amazon asked, “What if we could eliminate ETL entirely? That would be a world we would all love. This is our vision, what we’re calling a zero ETL future. And in this future, data integration is no longer a manual effort”, announcing support for a “zero-ETL” solution that tightly integrates Amazon Aurora with Amazon Redshift. Under that integration, within seconds of transactional data being written into Aurora, the data is available in Amazon Redshift. 

The benefits of an integration like this are obvious: No need to build and maintain complex data pipelines, no duplicate data storage (which can be expensive), and always up-to-date.

Now, an integration between two Amazon databases in itself is not enough to lead to the end of ETL alone, and there are reasons to be skeptical that a Zero ETL future would happen soon. 

But then again, Salesforce and Snowflake also announced a partnership to share customer data in real-time across systems without moving or copying data, which falls under the same general logic. Before that, Stripe had launched a data pipeline to help users sync payment data with Redshift and Snowflake. 

The concept of change data capture is not new, but it’s gaining steam. Google already supports change data capture in BigQuery. Azure Synapse does the same by pre-integrating Azure Data Factory. There is a rising generation of startups in the space like Estuary* and Upsolver. It seems that we’re heading towards a hybrid future where analytic platforms will blend in streaming, integration with data flow pipelines and Kafka PubSub feeds.

Reverse ETL vs. CDP

Another somewhat-in-the-weeds but fun-to-watch part of the landscape has been the tension between Reverse ETL (again, the process of taking data out of the warehouse and putting it back into SaaS and other applications) and Customer Data Platforms (products that aggregate customer data from multiple sources, run analytics on them like segmentation, and enable actions like marketing campaigns). 

Over the last year or so, the two categories started converging into one another.  

Reverse ETL companies presumably learned that just being a pipeline on top of a data warehouse wasn’t commanding enough wallet share from customers and that they needed to go further in providing value around customer data. Many Reverse ETL vendors now position themselves as CDP from a marketing standpoint.   

Meanwhile, CDP vendors learned that being another repository where customers needed to copy massive amounts of data was at odds with the general trend of centralization of data around the data warehouse (or lake or lakehouse). Therefore, CDP vendors started offering integration with the main data warehouse and lakehouse providers. See, for example, ActionIQ* launching HybridCompute, mParticle launching Warehouse Sync, or Segment introducing Reverse ETL capabilities. As they beef up their own reverse ETL capabilities, CDP companies are now starting to sell to a more technical audience of CIO and analytics teams, in addition to their historical buyers (CMOs).

Where does this leave Reverse ETL companies? One way they could evolve is to become more deeply integrated with the ETL providers, which we discussed above. Another way would be to further evolve towards becoming a CDP by adding analytics and orchestration modules.  

Data mesh, products, contracts: Dealing with organizational complexity

As just about any data practitioner knows firsthand: success with data is certainly a technical and product effort, but it also very much revolves around process and organizational issues.

In many organizations, the data stack looks like a mini-version of the MAD landscape. You end up with a variety of teams working on a variety of products. So how does it all work together? Who’s in charge of what?

A debate has been raging in data circles about how to best go about it. There are a lot of nuances and a lot of discussions with smart people disagree on, well, just about any part of it, but here’s a quick overview. 

We highlighted the data mesh as an emerging trend in the 2021 MAD landscape and it’s only been gaining traction since. The data mesh is a distributed, decentralized (not in the crypto sense) approach to managing data tools and teams. Note how it’s different from a data fabric – a more technical concept, basically a single framework to connect all data sources within the enterprise, regardless of where they’re physically located.

The data mesh leads to a concept of data products – which could be anything from a curated data set to an application or an API. The basic idea is that each team that creates the data product is fully responsible for it (including quality, uptime, etc.). Business units within the enterprise then consume the data product on a self-service basis. 

A related idea is data contracts: “API-like agreements between software engineers who own services and data consumers that understand how the business works in order to generate well-modeled, high-quality, trusted, real-time data.” There have been all sorts of fun debates about the concept. The essence of the discussion is whether data contracts only make sense in very large, very decentralized organizations, as opposed to 90% of smaller companies. 

Bonus: How will AI impact data infrastructure? 

With the current explosive progress in AI, here’s a fun question: Data infrastructure has certainly been powering AI, but will AI now impact data infrastructure?

Some data infrastructure providers have already been using AI for a while – see, for example, Anomalo leveraging ML to identify data quality issues in the data warehouse. But with the rise of Large Language Models, there’s a new interesting angle. In the same way LLMs can create conventional programming code, they can also generate SQL, the language of data analysts. The idea of enabling non-technical users to search analytical systems is not new, and various providers already support variations of it, see ThoughtSpot, Power BI or Tableau.  Here are some good pieces on the topic: LLM Implications on Analytics (and Analysts!) by Tristan Handy of dbt Labs and The Rapture and the Reckoning by Benn Stancil of Mode. 

MAD 2023, part IV: Trends in ML/AI

The excitement! The drama! The action!

Everybody is talking breathlessly about AI all of a sudden. OpenAI gets a $10B investment. Google is in Code Red. Sergey is coding again. Bill Gates says what’s been happening in AI in the last 12 months is “every bit as important as the PC or the internet.” Brand new startups are popping up (20 generative AI companies just in the winter ’23 YC batch). VCs are back to chasing pre-revenue startups at billions of valuation.

So what does it all mean? Is this one of those breakthrough moments that only happen every few decades? Or just the logical continuation of work that has been happening for many years? Are we in the early days of a true exponential acceleration? Or at the top of one of those hype cycles, as many in tech are desperate for the next big platform shift after social and mobile and the crypto headfake?

The answer to all those questions is… yes.

Let’s dig in:

• AI goes mainstream
• Generative AI becomes a household name
• The inevitable backlash
• [Big progress in reinforcement learning]
• [The emergence of a new AI political economy]
• [Big Tech has a head start over startups]
• [Are we getting closer to AGI?]

AI goes mainstream

It had been a wild ride in the world of AI throughout 2022, but what truly took things to a fever pitch was, of course, the public release of Open’s AI conversational bot, ChatGPT, on November 30, 2022. ChatGPT, a chatbot with an uncanny ability to mimic a human conversationalist, quickly became the fastest-growing product, well, ever.

For whoever was around then, the experience of first interacting with ChatGPT was reminiscent of the first time they interacted with Google in the late nineties. Wait, is it really that good? And that fast? How is this even possible? Or the iPhone when it first came out. Basically, a first glimpse into what feels like an exponential future. 

ChatGPT immediately took over every business meeting, conversation, dinner, and, most of all, every bit of social media. Screenshots of smart, amusing and occasionally wrong replies by ChatGPT became ubiquitous on Twitter. We all just had to chat about ChatGPT.

By January, ChatGPT had reached 100M users. A whole industry of overnight experts emerged on social media, with a never-ending bombardment of explainer threads coming to the rescue of anyone who had been struggling with ChatGPT (literally, no one asked) and ambitious TikTokers teaching us the ways of prompt engineering, meaning providing the kind of input that would elicit the best response from ChatGPT.  

After being exposed to a non-stop barrage of tweets on the topic, this was the sentiment:

ChatGPT continued to accumulate feats.  It passed the Bar.  It passed the US medical licensing exam. 

ChatGPT didn’t come out of nowhere. AI circles had been buzzing about GPT-3 since its release in June 2020, raving about a quality of text output that was so high that it was difficult to determine whether or not it was written by a human. But GPT-3 was provided as an API targeting developers, not the broad public. 

The release of ChatGPT (based on GPT 3.5) feels like the moment AI truly went mainstream in the collective consciousness.  

We are all routinely exposed to AI prowess in our everyday lives through voice assistants, auto-categorization of photos, using our faces to unlock our cell phones, or receiving calls from our banks after an AI system detected possible financial fraud.  But, beyond the fact that most people don’t realize that AI powers all of those capabilities and more, arguably, those feel like one-trick ponies.  

With ChatGPT, suddenly, you had the experience of interacting with something that felt like an all-encompassing intelligence.

The hype around ChatGPT is not just fun to talk about. It’s very consequential because it has forced the industry to react aggressively to it, unleashing, among other things, an epic battle for internet search. 

The exponential acceleration of generative AI 

But, of course, it’s not just ChatGPT. For anyone who was paying attention, the last few months saw a dizzying succession of groundbreaking announcements seemingly every day. With AI, you could now create audio, code, images, text and videos. 

What was at some point called synthetic media (a category in the 2021 MAD landscape) became widely known as generative AI: A term still so new that it does not have an entry in Wikipedia at the time of writing. 

The rise of generative AI has been several years in the making. Depending on how you look at it, it traces it roots back to deep learning (which is several decades old but dramatically accelerated after 2012) and the advent of generative Adversarial Networks (GAN) in 2014, led by Ian Goodfellow, under the supervision of his professor and Turing Award recipient, Yoshua Bengio. Its seminal moment, however, came barely five years ago, with the publication of the transformer (the “T” in GPT) architecture in 2017, by Google.

Coupled with rapid progress in data infrastructure, powerful hardware and a fundamentally collaborative, open source approach to research, the transformer architecture gave rise to the Large Language Model (LLM) phenomenon.

The concept of a language model itself is not particularly new.  A language model’s core function is to predict the next word in a sentence.

However, transformers brought a multimodal dimension to language models. There used to be separate architectures for computer vision, text and audio. With transformers, one general architecture can now gobble up all sorts of data, leading to an overall convergence in AI. 

In addition, the big change has been the ability to massively scale those models.  

OpenAI’s GPT models are a flavor of transformers that it trained on the Internet, starting in 2018. GPT-3, their third-generation LLM, is one of the most powerful models currently available. It can be fine-tuned for a wide range of tasks – language translation, text summarization, and more. GPT-4 is expected to be released sometime in 2024 and is rumored to be even more mind-blowing. (ChatGPT is based on GPT 3.5, a variant of GPT-3).

OpenAI also played a driving role in AI image generation. In early 2021, it released CLIP, an open source, multimodal, zero-shot model. Given an image and text descriptions, the model can predict the most relevant text description for that image without optimizing for a particular task.

OpenAI doubled down with DALL-E, an AI system that can create realistic images and art from a description in natural language. The particularly impressive second version, DALL-E 2, was broadly released to the public at the end of September 2022.

There are already multiple contenders vying to be the best text-to-image model. Midjourney, entered open beta in July 2022 (it’s currently only accessible through their Discord*).  Stable Diffusion, another impressive model, was released in August 2022.  It originated through the collaboration of several entities, in particular Stability AI, CompVis LMU, and Runway ML. It offers the distinction of being open source, which DALL-E 2 and Midjourney are not.

Those developments are not even close to the exponential acceleration of AI releases that occurred since the middle of 2022. 

In September 2022, OpenAI released Whisper, an automatic speech recognition (ASR) system that enables transcription in multiple languages as well as translation from those languages into English. Also in September 2022, MetaAI released Make-A-Video, an AI system that generates videos from text.

In October 2022, CSM (Common Sense Machines) released CommonSim-1, a model to create 3D worlds.

In November 2022, MetaAI released CICERO, the first AI to play the strategy game Diplomacy at a human level, described as “a step forward in human-AI interactions with AI that can engage and compete with people in gameplay using strategic reasoning and natural language.”

In January 2023, Google Research announced MusicLM, “a model generating high-fidelity music from text descriptions such as “a calming violin melody backed by a distorted guitar riff.”

Another particularly fertile area for generative AI has been the creation of code.

In 2021, OpenAI released Codex, a model that translates natural language into code. You can use codex for tasks like “turning comments into code, rewriting code for efficiency, or completing your next line in context.” Codex is based on GPT-3 and was also trained on 54 million GitHub repositories. In turn, GitHub Copilot uses Codex to suggest code right from the editor.

In turn, Google’s DeepMind released Alphacode in February 2022 and Salesforce released CodeGen in March 2022.  Huawei introduced PanGu-Coder in July 2022. 

The inevitable backlash

The exponential acceleration in AI progress over the last few months has taken most people by surprise. It is a clear case where technology is way ahead of where we are as humans in terms of society, politics, legal framework and ethics. For all the excitement, it was received with horror by some and we are just in the early days of figuring out how to handle this massive burst of innovation and its consequences. 

ChatGPT was pretty much immediately banned by some schools, AI conferences (the irony!) and programmer websites. Stable Diffusion was misused to create an NSFW porn generator, Unstable Diffusion, later shut down on Kickstarter.  There are allegations of exploitation of Kenyan workers involved in the data labeling process. Microsoft/GitHub is getting sued for IP violation when training Copilot, accused of killing open source communities.  Stability AI is getting sued by Getty for copyright infringement.  Midjourney might be next (Meta is partnering with Shutterstock to avoid this issue). When an A.I.-generated work, “Théâtre d’Opéra Spatial,” took first place in the digital category at the Colorado State Fair, artists around the world were up in arms. 

AI and jobs

A lot of people’s reaction when confronted with the power of generative AI is that it will kill jobs. The common wisdom in years past was that AI would gradually automate the most boring and repetitive jobs. AI would kill creative jobs last because creativity is the most quintessentially human trait. But here we are, with generative AI going straight after creative pursuits.   

Artists are learning to co-create with AI.  Many are realizing that there’s a different kind of skill involved. Jason Allen, the creator of Théâtre d’Opéra Spatial, explains that he spent 80 hours and created 900 images before getting to the perfect combination. 

Similarly, coders are figuring out how to work alongside Copilot. AI leader, Andrej Karpathy, says Copilot already writes 80% of his code. Early research seems to indicate significant improvements in developer productivity and happiness. It seems that we’re evolving towards a co-working model where AI models work alongside humans as “pair programmers” or “pair artists.”  

Perhaps AI will lead to the creation of new jobs. There’s already a marketplace for selling high-quality text prompts.

AI bias

A serious strike against generative AI is that it is biased and possibly toxic. Given that AI reflects its training dataset, and considering GPT and others were trained on the highly biased and toxic Internet, it’s no surprise that this would happen. 

Early research has found that image generation models, like Stable Diffusion and DALL-E, not only perpetuate but also amplify demographic stereotypes.

At the time of writing, there is a controversy in conservative circles that ChatGPT is painfully woke. 

AI disinformation 

Another inevitable question is all the nefarious things that can be done with such a powerful new tool. New research shows AI’s ability to simulate reactions from particular human groups, which could unleash another level in information warfare.

Gary Marcus warns us about AI’s Jurassic Park moment – how disinformation networks would take advantage of ChatGPT, “attacking social media and crafting fake websites at a volume we have never seen before.”

AI platforms are moving promptly to help fight back, in particular by detecting what was written by a human vs. what was written by an AI.  OpenAI just launched a new classifier to do that, which is beating the state of the art in detecting AI-generated text.  

Is AI content just… boring?

Another strike against generative AI is that it could be mostly underwhelming. 

Some commentators worry about an avalanche of uninteresting, formulaic content meant to help with SEO or demonstrate shallow expertise, not dissimilarly from what content farms (a la Demand Media) used to do.

As Jack Clark pouts in his OpenAI newsletter: “Are we building these models to enrich our own experience, or will these models ultimately be used to slice and dice up human creativity and repackage and commoditize it? Will these models ultimately enforce a kind of cultural homogeneity acting as an anchor forever stuck in the past? Or could these models play their own part in a new kind of sampling and remix culture for music?”

AI hallucination

Finally, perhaps the biggest strike against generative AI is that it is often just wrong. 

ChatGPT, in particular, is known for “hallucinating,” meaning making up facts while conveying them with utter self-confidence in its answers.

Leaders in AI have been very explicit about it, like OpenAI’s CEO Sam Altman here: 

The big companies are well aware of the risk.

MetaAI introduced Galactica, a model designed to assist scientists, in November 2022 but pulled it after three days. The model generated both convincing scientific content and convincing (and occasionally racist) content. 

Google kept its LaMBDA model very private, available to only a small group of people through AI Test Kitchen, an experimental app. The genius of Microsoft working with OpenAI as an outsourced research arm was that OpenAI, as a startup, could take risks that Microsoft could not. One can assume that Microsoft was still reeling from the Tay disaster in 2016.

However, Microsoft was forced by competition (or could not resist the temptation) to open Pandora’s box and add GPT to its Bing search engine. That did not go as well as it could have, with Bing threatening users or declaring their love to them. 

Subsequently, Google also rushed to market its own ChatGPT competitor, the interestingly named Bard. This did not go well either, and Google lost $100B in market capitalization after Bard made factual errors in its first demo.

The business of AI: Big Tech has a head start over startups

The question on everyone’s minds in venture and startup circles: what is the business opportunity? The recent history of technology has seen a major platform shift every 15 years or so for the last few decades: the mainframe, the PC, the internet and mobile.  Many thought crypto and blockchain architecture was the next big shift but, at a minimum, the jury is out on that one for now. 

Is generative AI that once-every-15-years kind of generational opportunity that is about to unleash a massive new wave of startups (and funding opportunities for VCs)? Let’s look into some of the key questions.

Will incumbents own the market?

The success story in Silicon Valley lore goes something like this: big incumbent owns a large market but gets entitled and lazy; little startup comes up with a 10x better technology; against the odds and through great execution (and judicious from the VCs on the board, of course), little startup hits hyper-growth, becomes big and overtakes the big incumbent.

The issue in AI is that little startups are facing a very specific type of incumbents – the world’s biggest technology companies, including Alphabet/Google, Microsoft, Meta/Facebook and Amazon/AWS.  

Not only are those incumbents not “lazy,” but in many ways, they’ve been leading the charge in innovation in AI. Google thought of itself as an AI company from the very beginning (“Artificial intelligence would be the ultimate version of Google… that is basically what we work on,” said Larry Page in 2000). The company produced many key innovations in AI, including transformers, as mentioned, Tensorflow and the Tensor Processing Units (TPU). Meta/Facebook We talked about how Transformers came from Google, but that’s just one of the many innovations that the company has released over the years.  Meta/Facebook created PyTorch, one of the most important and used machine learning frameworks. Amazon, Apple, Microsoft, Netflix have all produced groundbreaking work. 

Incumbents also have some of the very best research labs, experienced machine learning engineers, massive amounts of data, tremendous processing power and enormous distribution and branding power. 

And finally, AI is likely to become even more of a top priority as it is becoming a major battleground. As mentioned earlier, Google and Microsoft are now engaged in an epic battle in search, with Microsoft viewing GPT as an opportunity to breathe new life into Bing and Google, considering it a potentially life-threatening alert. 

Meta/Facebook has made a huge bet in a very different area – the metaverse. That bet continues to prove to be very controversial.  Meanwhile, it’s sitting on some of the best AI talent and technology in the world.  How long until it reverses course and starts doubling or tripling down on AI?

Is AI just a feature?

Beyond Bing, Microsoft quickly rolled out GPT in Teams. Notion launched NotionAI, a new GPT-3-powered writing assistant.  Quora launched Poe, its own AI chatbot. Customer service leaders Intercom and Ada* announced GPT-powered features. 

How quickly and seemingly easily companies are rolling out AI-powered features seems to indicate that AI is going to be everywhere soon. In prior platform shifts, a big part of the story was that every company out there adopted the new platform: Businesses became internet-enabled, everyone built a mobile app, etc. 

We don’t expect anything different to happen here. We’ve long argued in prior posts that the success of data and AI technologies is that they eventually will become ubiquitous and disappear in the background. It’s the ransom of success for enabling technologies to become invisible. 

What are the opportunities for startups?

However, as history has shown time and again, don’t discount startups.  Give them a technology breakthrough, and entrepreneurs will find a way to build great companies.  

Yes, when mobile appeared, all companies became mobile-enabled.  However, founders built great startups that could not have existed without the mobile platform shift – Uber being the most obvious example. 

Who will be the Uber of generative AI?

The new generation of AI Labs is perhaps building the AWS, rather than Uber, of generative AI.  OpenAI, Anthropic, Stability AI, Adept, Midjourney and others are building broad horizontal platforms upon which many applications are already being created.  It is an expensive business, as building large language models is extremely resource intensive, although perhaps costs are going to drop rapidly. The business model of those platforms is still being worked out.  OpenAI launched ChatGPT Plus, a paying premium version of ChatGPT. Stability AI plans on monetizing its platform by charging for customer-specific versions. 

There’s been an explosion of new startups leveraging GPT, in particular, for all sorts of generative tasks, from creating code to marketing copy to videos.  Many are derided as being a “thin layer” on top of GPT.  There’s some truth to that, and their defensibility is unclear.  But perhaps that’s the wrong question to ask.  Perhaps those companies are just the next generation of software rather than AI companies.  As they build more functionality around things like workflow and collaboration on top of the core AI engine, they will be no more, but also no less, defensible than your average SaaS company. 

We believe that there are many opportunities to build great companies: vertical-specific or task-specific companies that will intelligently leverage generative AI for what it is good at. AI-first companies that will develop their own models for tasks that are not generative in nature. LLM-ops companies that will provide the necessary infrastructure. And so many more. 

This next wave is just getting started, and we can’t wait to see what happens.

Matt Turck is a VC at FirstMark, where he focuses on SaaS, cloud, data, ML/AI, and infrastructure investments. Matt also organizes Data Driven NYC, the largest data community in the U.S.

This story originally appeared on Mattturck.com. Copyright 2023