AI

There’s now an open source alternative to ChatGPT, but good luck running it

Comment

Image of a person typing on a computer.
Image Credits: Basak Gurbuz Derman (opens in a new window) / Getty Images

The first open source equivalent of OpenAI’s ChatGPT has arrived, but good luck running it on your laptop — or at all.

This week, Philip Wang, the developer responsible for reverse-engineering closed-sourced AI systems including Meta’s Make-A-Video, released PaLM + RLHF, a text-generating model that behaves similarly to ChatGPT. The system combines PaLM, a large language model from Google, and a technique called Reinforcement Learning with Human Feedback — RLHF, for short — to create a system that can accomplish pretty much any task that ChatGPT can, including drafting emails and suggesting computer code.

But PaLM + RLHF isn’t pre-trained. That is to say, the system hasn’t been trained on the example data from the web necessary for it to actually work. Downloading PaLM + RLHF won’t magically install a ChatGPT-like experience — that would require compiling gigabytes of text from which the model can learn and finding hardware beefy enough to handle the training workload.

Like ChatGPT, PaLM + RLHF is essentially a statistical tool to predict words. When fed an enormous number of examples from training data — e.g., posts from Reddit, news articles and e-books — PaLM + RLHF learns how likely words are to occur based on patterns like the semantic context of surrounding text.

ChatGPT and PaLM + RLHF share a special sauce in Reinforcement Learning with Human Feedback, a technique that aims to better align language models with what users wish them to accomplish. RLHF involves training a language model — in PaLM + RLHF’s case, PaLM — and fine-tuning it on a dataset that includes prompts (e.g., “Explain machine learning to a six-year-old”) paired with what human volunteers expect the model to say (e.g., “Machine learning is a form of AI…”). The aforementioned prompts are then fed to the fine-tuned model, which generates several responses, and the volunteers rank all the responses from best to worst. Finally, the rankings are used to train a “reward model” that takes the original model’s responses and sorts them in order of preference, filtering for the top answers to a given prompt.

It’s an expensive process, collecting the training data. And training itself isn’t cheap. PaLM is 540 billion parameters in size, “parameters” referring to the parts of the language model learned from the training data. A 2020 study pegged the expenses for developing a text-generating model with only 1.5 billion parameters at as much as $1.6 million. And to train the open source model Bloom, which has 176 billion parameters, it took three months using 384 Nvidia A100 GPUs; a single A100 costs thousands of dollars.

Running a trained model of PaLM + RLHF’s size isn’t trivial, either. Bloom requires a dedicated PC with around eight A100 GPUs. Cloud alternatives are pricey, with back-of-the-envelope math finding the cost of running OpenAI’s text-generating GPT-3 — which has around 175 billion parameters — on a single Amazon Web Services instance to be around $87,000 per year.

Sebastian Raschka, an AI researcher, points out in a LinkedIn post about PaLM + RLHF that scaling up the necessary dev workflows could prove to be a challenge as well. “Even if someone provides you with 500 GPUs to train this model, you still need to have to deal with infrastructure and have a software framework that can handle that,” he said. “It’s obviously possible, but it’s a big effort at the moment (of course, we are developing frameworks to make that simpler, but it’s still not trivial, yet).”

That’s all to say that PaLM + RLHF isn’t going to replace ChatGPT today — unless a well-funded venture (or person) goes to the trouble of training and making it available publicly.

In better news, several other efforts to replicate ChatGPT are progressing at a fast clip, including one led by a research group called CarperAI. In partnership with the open AI research organization EleutherAI and startups Scale AI and Hugging Face, CarperAI plans to release the first ready-to-run, ChatGPT-like AI model trained with human feedback.

LAION, the nonprofit that supplied the initial dataset used to train Stable Diffusion, is also spearheading a project to replicate ChatGPT using the newest machine learning techniques. Ambitiously, LAION aims to build an “assistant of the future” — one that not only writes emails and cover letters but “does meaningful work, uses APIs, dynamically researches information and much more.” It’s in the early stages. But a GitHub page with resources for the project went live a few weeks ago.

More TechCrunch

Hiya, folks, welcome to TechCrunch’s regular AI newsletter. Last Sunday, President Joe Biden announced that he no longer plans to seek reelection, instead offering his “full endorsement” of VP Kamala…

This Week in AI: How Kamala Harris might regulate AI

But the fate of many generative AI businesses — even the best-funded ones — looks murky.

VCs are still pouring billions into generative AI startups

Thousands of stories have been written about former NFL quarterback and civil rights activist Colin Kaepernick. If anyone knows a thing or two about losing control of your own narrative,…

Colin Kaepernick lost control of his story. Now he wants to help creators own theirs

Several people who received the CrowdStrike offer found that the gift card didn’t work, while others got an error saying the voucher had been canceled.

CrowdStrike offers a $10 apology gift card to say sorry for outage

TikTok Lite, a low-bandwidth version of the video platform popular across Africa, Asia and Latin America, is exposing users to harmful content because of its lack of safety features compared…

TikTok Lite exposes users to harmful content, say Mozilla researchers

If the models continue eating each other’s data, perhaps without even knowing it, they’ll progressively get weirder and dumber until they collapse.

‘Model collapse’: Scientists warn against letting AI eat its own tail

Astranis has fully funded its next-generation satellite program, called Omega, after closing its $200 million Series D round, the company said Wednesday.  “This next satellite is really the milestone into…

Astranis is set to build Omega constellation after $200M Series D

Reworkd’s founders went viral on GitHub last year with AgentGPT, a free tool to build AI agents that acquired more than 100,000 daily users in a week. This earned them…

After AgentGPT’s success, Reworkd pivots to web-scraping AI agents

We’re so excited to announce that we’ve added a dedicated AI Stage presented by Google Cloud to TechCrunch Disrupt 2024. It joins Fintech, SaaS and Space as the other industry-focused…

Announcing the agenda for the AI Stage at TechCrunch Disrupt 2024

The firm has numerous legs to it, ranging from a venture studio to standard funds, where it does everything from co-founding companies to deploying capital.

CityRock launches second fund to back founders from diverse backgrounds

Since launching xAI last year, Elon Musk has been using X as a sandbox to test some of the Grok model’s AI capabilities. Beyond the basic chatbot, X uses the…

X launches underwhelming Grok-powered ‘More About This Account’ feature

Lakera, a Swiss startup that’s building technology to protect generative AI applications from malicious prompts and other threats, has raised $20 million in a Series A round led by European…

Lakera, which protects enterprises from LLM vulnerabilities, raises $20M

Alongside a slew of announcements for Play—such as AI-powered app comparisons and a feature that bundles similar apps—Google has introduced new “Curated Spaces,” hubs dedicated to specific topics. Announced Wednesday,…

Google Play gets ‘Comics’ feature for manga readers in Japan

Farmers have got to do something about pests. But nobody really likes the idea of using more chemical pesticides. Thomas Laurent’s company, Micropep, thinks the answer might already be in…

Micropep taps tiny proteins to make pesticides safer

Play Store is getting AI-powered app comparisons, automatically organized categories for similar apps, dedicated hubs for content, data personalization controls, support for playing multiple mobile games on PCs, and more…

Google adds AI-powered comparisons, collections and more data controls to Play Store

Vanta, a trust management platform that helps businesses automate much of their security and compliance processes, today announced that it has raised a $150 million Series C funding round led…

Vanta raises $150M Series C, now valued at $2.45B

The Overture Maps Foundation is today releasing data sets for 2.3B building “footprints” globally, 54M notable places of interest, a visual overlay of “boundaries,” and land and water features such…

Backed by Microsoft, AWS and Meta, the Overture Maps Foundation launches its first open map data sets

The startup is not disclosing its valuation, but sources close to the company say the figure is just under $400 million post-money.

Dazz snaps up $50M for AI-based, automated cloud security remediation

The outcome of the Spanish authority’s probe could take up to two years to complete, and leave Apple on the hook for fines in the billions.

Apple’s App Store hit with antitrust probe in Spain

Proton’s first cryptocurrency product is a wallet called Proton Wallet that’s designed to make it easier to get started with bitcoin.

Proton releases a self-custody bitcoin wallet

Dental care is a necessity, yet many patients lack confidence in their dentists’ ability to provide accurate diagnoses and appropriate treatments. Some dentists over treat patients, leading to unnecessary expenses,…

Pearl raises $58M to help dentists make better diagnoses using AI 

Exoticca’s platform connects flights, hotels, meals, transfers, transportation and more, plus the local companies at the destinations.

Spanish startup Exoticca raises a €60M Series D for its tour packages platform

Content creators are busy people. Most spend more than 20 hours a week creating new content for their respective corners of the web. That doesn’t leave much time for audience…

Mark Zuckerberg imagines content creators making AI clones of themselves

Elon Musk says he will show off Tesla’s purpose-built “robotaxi” prototype during an event October 10, after scrapping a previous plan to reveal it August 8. Musk said Tesla will…

Elon Musk sets new date for Tesla robotaxi reveal, calls everything beyond autonomy ‘noise’

Alphabet will spend an additional $5 billion on its self-driving subsidiary, Waymo, over the next few years, according to Ruth Porat, the company’s chief financial officer. Porat announced the commitment…

Alphabet to invest another $5B into Waymo

There is no fool proof way to prevent a buggy update like CrowdStrike’s, but there are best practices that could mitigate the fallout.

How to prevent your software update from being the next CrowdStrike

Spotify CEO Daniel Ek says the streaming service is still in the “early days” of its plans to bring hi-fi support to the platform. During the company’s earnings call on…

Spotify CEO says company is in ‘early days’ of hi-fi audio plans

Featured Article

A comprehensive list of 2024 tech layoffs

The tech layoff wave is still going strong in 2024. Following significant workforce reductions in 2022 and 2023, this year has already seen 60,000 job cuts across 254 companies, according to independent layoffs tracker Layoffs.fyi. Companies like Tesla, Amazon, Google, TikTok, Snap and Microsoft have conducted sizable layoffs in the…

A comprehensive list of 2024 tech layoffs

Tesla was not the first company to begin working on a humanoid form factor, but while being the first to market does carry weight in this high-tech space, we’re at…

Elon Musk sets 2026 Optimus sale date. Here’s where other humanoid robots stand.

Harvey, a startup building what it describes as an AI-powered “copilot” for lawyers, has raised $100 million in a Series C round led by GV, Google’s corporate venture arm. The…

OpenAI-backed legal tech startup Harvey raises $100M