AI

AI training data has a price tag that only Big Tech can afford

Comment

Binary code in blue with little yellow locks in between to illustrate data protection.
Image Credits: Peresmeh / Getty Images

Data is at the heart of today’s advanced AI systems, but it’s costing more and more — making it out of reach for all but the wealthiest tech companies.

Last year, James Betker, a researcher at OpenAI, penned a post on his personal blog about the nature of generative AI models and the datasets on which they’re trained. In it, Betker claimed that training data — not a model’s design, architecture or any other characteristic — was the key to increasingly sophisticated, capable AI systems.

“Trained on the same data set for long enough, pretty much every model converges to the same point,” Betker wrote.

Is Betker right? Is training data the biggest determiner of what a model can do, whether it’s answer a question, draw human hands, or generate a realistic cityscape?

It’s certainly plausible.

Statistical machines

Generative AI systems are basically probabilistic models — a huge pile of statistics. They guess based on vast amounts of examples which data makes the most “sense” to place where (e.g., the word “go” before “to the market” in the sentence “I go to the market”). It seems intuitive, then, that the more examples a model has to go on, the better the performance of models trained on those examples.

“It does seem like the performance gains are coming from data,” Kyle Lo, a senior applied research scientist at the Allen Institute for AI (AI2), a AI research nonprofit, told TechCrunch, “at least once you have a stable training setup.”

Lo gave the example of Meta’s Llama 3, a text-generating model released earlier this year, which outperforms AI2’s own OLMo model despite being architecturally very similar. Llama 3 was trained on significantly more data than OLMo, which Lo believes explains its superiority on many popular AI benchmarks.

(I’ll point out here that the benchmarks in wide use in the AI industry today aren’t necessarily the best gauge of a model’s performance, but outside of qualitative tests like our own, they’re one of the few measures we have to go on.)

That’s not to suggest that training on exponentially larger datasets is a sure-fire path to exponentially better models. Models operate on a “garbage in, garbage out” paradigm, Lo notes, and so data curation and quality matter a great deal, perhaps more than sheer quantity.

“It is possible that a small model with carefully designed data outperforms a large model,” he added. “For example, Falcon 180B, a large model, is ranked 63rd on the LMSYS benchmark, while Llama 2 13B, a much smaller model, is ranked 56th.”

In an interview with TechCrunch last October, OpenAI researcher Gabriel Goh said that higher-quality annotations contributed enormously to the enhanced image quality in DALL-E 3, OpenAI’s text-to-image model, over its predecessor DALL-E 2. “I think this is the main source of the improvements,” he said. “The text annotations are a lot better than they were [with DALL-E 2] — it’s not even comparable.”

Many AI models, including DALL-E 3 and DALL-E 2, are trained by having human annotators label data so that a model can learn to associate those labels with other, observed characteristics of that data. For example, a model that’s fed lots of cat pictures with annotations for each breed will eventually “learn” to associate terms like bobtail and shorthair with their distinctive visual traits.

Bad behavior

Experts like Lo worry that the growing emphasis on large, high-quality training datasets will centralize AI development into the few players with billion-dollar budgets that can afford to acquire these sets. Major innovation in synthetic data or fundamental architecture could disrupt the status quo, but neither appear to be on the near horizon.

“Overall, entities governing content that’s potentially useful for AI development are incentivized to lock up their materials,” Lo said. “And as access to data closes up, we’re basically blessing a few early movers on data acquisition and pulling up the ladder so nobody else can get access to data to catch up.”

Indeed, where the race to scoop up more training data hasn’t led to unethical (and perhaps even illegal) behavior like secretly aggregating copyrighted content, it has rewarded tech giants with deep pockets to spend on data licensing.

Generative AI models such as OpenAI’s are trained mostly on images, text, audio, videos and other data — some copyrighted — sourced from public web pages (including, problematically, AI-generated ones). The OpenAIs of the world assert that fair use shields them from legal reprisal. Many rights holders disagree — but, at least for now, they can’t do much to prevent this practice.

There are many, many examples of generative AI vendors acquiring massive datasets through questionable means in order to train their models. OpenAI reportedly transcribed more than a million hours of YouTube videos without YouTube’s blessing — or the blessing of creators — to feed to its flagship model GPT-4. Google recently broadened its terms of service in part to be able to tap public Google Docs, restaurant reviews on Google Maps and other online material for its AI products. And Meta is said to have considered risking lawsuits to train its models on IP-protected content.

Meanwhile, companies large and small are relying on workers in third-world countries paid only a few dollars per hour to create annotations for training sets. Some of these annotators — employed by mammoth startups like Scale AI — work literal days on end to complete tasks that expose them to graphic depictions of violence and bloodshed without any benefits or guarantees of future gigs.

Growing cost

In other words, even the more aboveboard data deals aren’t exactly fostering an open and equitable generative AI ecosystem.

OpenAI has spent hundreds of millions of dollars licensing content from news publishers, stock media libraries and more to train its AI models — a budget far beyond that of most academic research groups, nonprofits and startups. Meta has gone so far as to weigh acquiring the publisher Simon & Schuster for the rights to e-book excerpts (ultimately, Simon & Schuster sold to private equity firm KKR for $1.62 billion in 2023).

With the market for AI training data expected to grow from roughly $2.5 billion now to close to $30 billion within a decade, data brokers and platforms are rushing to charge top dollar — in some cases over the objections of their user bases.

Stock media library Shutterstock has inked deals with AI vendors ranging from $25 million to $50 million, while Reddit claims to have made hundreds of millions from licensing data to orgs such as Google and OpenAI. Few platforms with abundant data accumulated organically over the years haven’t signed agreements with generative AI developers, it seems — from Photobucket to Tumblr to Q&A site Stack Overflow.

It’s the platforms’ data to sell — at least depending on which legal arguments you believe. But in most cases, users aren’t seeing a dime of the profits. And it’s harming the wider AI research community.

“Smaller players won’t be able to afford these data licenses, and therefore won’t be able to develop or study AI models,” Lo said. “I worry this could lead to a lack of independent scrutiny of AI development practices.”

Independent efforts

If there’s a ray of sunshine through the gloom, it’s the few independent, not-for-profit efforts to create massive datasets anyone can use to train a generative AI model.

EleutherAI, a grassroots nonprofit research group that began as a loose-knit Discord collective in 2020, is working with the University of Toronto, AI2 and independent researchers to create The Pile v2, a set of billions of text passages primarily sourced from the public domain.

In April, AI startup Hugging Face released FineWeb, a filtered version of the Common Crawl — the eponymous dataset maintained by the nonprofit Common Crawl, composed of billions upon billions of web pages — that Hugging Face claims improves model performance on many benchmarks.

A few efforts to release open training datasets, like the group LAION’s image sets, have run up against copyright, data privacy and other, equally serious ethical and legal challenges. But some of the more dedicated data curators have pledged to do better. The Pile v2, for example, removes problematic copyrighted material found in its progenitor dataset, The Pile.

The question is whether any of these open efforts can hope to maintain pace with Big Tech. As long as data collection and curation remains a matter of resources, the answer is likely no — at least not until some research breakthrough levels the playing field.

More TechCrunch

Employees at Bethesda Game Studios — the Microsoft-owned game developer that produces the Elder Scrolls and Fallout franchises — are joining the Communication Workers of America. Quality assurance testers at…

Bethesda Game Studios employees form a ‘wall-to-wall’ union

This week saw one of the most widespread IT disruptions in recent years linked to a faulty software update from popular cybersecurity firm CrowdStrike. Businesses across the world reported IT…

CrowdStrike’s update fail causes global outages and travel chaos

Alphabet, the parent company of Google, is in advanced talks to acquire cybersecurity startup Wiz for $23 billion, the Wall Street Journal reported on Sunday. TechCrunch’s sources heard similar and…

Unpacking how Alphabet’s rumored Wiz acquisition could affect VC

Around 8.5 million devices — less than 1 percent Windows machines globally — were affected by the recent CrowdStrike outage, according to a Microsoft blog post by David Weston, the…

Microsoft says 8.5M Windows devices were affected by CrowdStrike outage

Featured Article

Some Black startup founders feel betrayed by Ben Horowitz’s support for Trump

Trump is an advocate for a number of policies that could be harmful to people of color.

Some Black startup founders feel betrayed by Ben Horowitz’s support for Trump

Featured Article

Strava’s next chapter: New CEO talks AI, inclusivity, and why ‘dark mode’ took so long

TechCrunch sat down with Strava’s new CEO in London for a wide-ranging interview, delving into what the company is prioritizing, and what we can expect in the future as the company embarks on its “next chapter.”

Strava’s next chapter: New CEO talks AI, inclusivity, and why ‘dark mode’ took so long

Featured Article

Lavish parties and moral dilemmas: 4 days with Silicon Valley’s MAGA elite at the RNC

All week at the RNC, I saw an event defined by Silicon Valley. But I also saw the tech elite experience flashes of discordance.

Lavish parties and moral dilemmas: 4 days with Silicon Valley’s MAGA elite at the RNC

Featured Article

Tracking the EV battery factory construction boom across North America

A wave of automakers and battery makers — foreign and domestic — have pledged to produce North American–made batteries before 2030.

Tracking the EV battery factory construction boom across North America

Featured Article

Faulty CrowdStrike update causes major global IT outage, taking out banks, airlines and businesses globally

Security giant CrowdStrike said the outage was not caused by a cyberattack, as businesses anticipate widespread disruption.

Faulty CrowdStrike update causes major global IT outage, taking out banks, airlines and businesses globally

CISA confirmed the CrowdStrike outage was not caused by a cyberattack, but urged caution as malicious hackers exploit the situation.

US cyber agency CISA says malicious hackers are ‘taking advantage’ of CrowdStrike outage

The global outage is a perfect reminder how much of the world relies on technological infrastructure.

These startups are trying to prevent another CrowdStrike-like outage, according to VCs

The CrowdStrike outage that hit early Friday morning and knocked out computers running Microsoft Windows has grounded flights globally. Major U.S. airlines including United Airlines, American Airlines and Delta Air…

CrowdStrike outage: How your plane, train and automobile travel may be affected

Prior to the ban, Trump’s team used his channel to broadcast some of his campaigns. With the ban now lifted, his channel can resume doing so.

Twitch reinstates Trump’s account ahead of the 2024 presidential election

This week, Google is in discussions to pay $23 billion for cloud security startup Wiz, SoftBank acquires Graphcore, and more.

M&A activity heats up with Wiz, Graphcore, etc.

CrowdStrike competes with a number of vendors, including SentinelOne and Palo Alto Networks but also Microsoft, Trellix, Trend Micro and Sophos, in the endpoint security market.

CrowdStrike’s rivals stand to benefit from its update fail debacle

The IT outage may have an unexpected effect on the climate: clearer skies and maybe lower temperatures this evening

CrowdStrike chaos leads to grounded aircraft — and maybe an unusual weather effect

There’s a man in Florida right now who wants to propose to his girlfriend while they’re on a beach vacation. He couldn’t get the engagement ring before he flew down…

The CrowdStrike outage is a plot point in a rom-com 

Here’s everything you need to know so far about the global outages caused by CrowdStrike’s buggy software update.

What we know about CrowdStrike’s update fail that’s causing global outages and travel chaos

This serves as an example for how easy it is to spread inaccurate information online during a time of immense global confusion and panic.

From the Sphere to false cyberattack claims, misinformation runs rampant amid CrowdStrike outage

Today is the final chance to save up to $800 on TechCrunch Disrupt 2024 tickets. Disrupt Deal Days event will end tonight at 11:59 p.m. PT. Don’t miss out on…

Last chance today: Secure major savings for TechCrunch Disrupt 2024!

Indian fintech Paytm’s struggles won’t seem to end. The company on Friday reported that its revenue declined by 36% and its loss more than doubled in the first quarter as…

Paytm loss widens and revenue shrinks as it grapples with regulatory clampdown

J. Michael Cline, the co-founder of Fandango and multiple other startups over his multi-decade career, died after falling from a Manhattan hotel, New York’s Deputy Commissioner of Public Information tells…

Fandango founder dies in fall from Manhattan skyscraper

Venture capital giant a16z fixed a security vulnerability in one of the firm’s websites after being warned by a security researcher.

Researcher finds flaw in a16z website that exposed some company data

Apple on Thursday announced its upcoming lineup of immersive video content for the Vision Pro. The list includes behind-the-scenes footage of the 2024 NBA All-Star Weekend, an immersive performance by…

Apple Vision Pro debuts immersive content featuring NBA players, The Weeknd and more

Biden centering Musk in his campaign is a notable escalation, considering he spent most of his presidency seemingly pretending the billionaire didn’t exist.

Elon Musk is now a villain in Joe Biden’s presidential campaign

Waymo would need a ground transportation permit to operate at SFO, which has yet to be approved.

Waymo wants to bring robotaxis to SFO, emails show

When Tade Oyerinde first set out to fundraise for his startup, Campus, a fully accredited online community college, it was incredibly difficult. VCs have backed for-profit education companies in the…

Why it made sense for an online community college to raise venture capital

Canadian private equity firm PartnerOne paid $28.2 million for HeadSpin, a mobile app testing startup whose founder was sentenced for fraud earlier this year, according to documents viewed by TechCrunch.…

PE firm PartnerOne paid $28M for HeadSpin, a fraction of its $1.1B valuation set by ICONIQ and Dell Technologies Capital

Meta has suspended the use of its AI assistant after Brazil’s National Data Protection Authority (ANPD) banned the company from training its AI models on personal data from Brazilians. The…

Meta puts a halt to training its generative AI tools in Brazil 

ChatGPT, OpenAI’s text-generating AI chatbot, has taken the world by storm since its launch in November 2022. What started as a tool to hyper-charge productivity through writing essays and code…

ChatGPT: Everything you need to know about the AI-powered chatbot