Welcome to our weekly roundup of human-crafted AI information.
This week OpenAI handed out API goodies.
California’s AI security invoice bought killed.
And NVIDIA stunned us with a strong open mannequin.
Let’s dig in.
Right here come the brokers
OpenAI didn’t announce any new fashions (or Sora) at its Dev Day occasion, however builders have been excited over new API options. The Realtime API might be a game-changer for making smarter functions that talk with customers and even act as brokers.
There have been rumors about OpenAI going the “for-profit” route and awarding Sam Altman billions of {dollars} in fairness however Altman dismissed these. Even so, the corporate is on the drive for extra funding they usually’ll count on some return for his or her money.
Apple has built-in OpenAI’s fashions into its gadgets however dropped out of the most recent funding spherical for OpenAI which is predicted to lift roughly $6.5 billion.
We’re unsure why Apple doesn’t need a piece of the OpenAI pie however it might need one thing to do with new developments with its Apple Intelligence. Or perhaps it’s associated to Sam Altman’s demand for exclusivity.
Inform me you don’t have any moat with out telling me you don’t have any moat. pic.twitter.com/3I18MosvOg
— Pedro Domingos (@pmddomingos) October 3, 2024
Kill invoice
Gavin Newsom needed to resolve between placing a security rev limiter on AI builders or letting them go full steam forward. Ultimately, he determined to veto California’s SB 1047 AI security invoice and provided some attention-grabbing the reason why.
Are we actually at some extent the place we face real AI dangers but?
Properly that escalated shortly https://t.co/xhZCITRJjE pic.twitter.com/aLZn4blS8G
— AI Notkilleveryoneism Memes ⏸️ (@AISafetyMemes) September 30, 2024
Newsom has signed a spread of AI payments over the past month associated to deepfakes, AI watermarking, baby security, performers’ AI rights, and election misinformation. Final week he signed AB 2013 which can actually shake issues up for LLM creators.
The invoice says that on or earlier than January 1, 2026, builders might want to present a high-level abstract of the coaching dataset of any fashions produced from January 1, 2022 onward if the mannequin is made obtainable in California. A few of these necessities may air some soiled secrets and techniques.
Extra EU AI regs
The EU is clearly much more involved about AI security than the remainder of the world. Both that or they simply get pleasure from writing and passing laws. This week they kickstarted a venture to write down an AI code of apply to try to steadiness innovation & security.
Once you see who heads up the security technical group you’ll have a good suggestion of which manner they’ll be leaning.
Liquid basis fashions
Transformer fashions are what gave us ChatGPT however there’s been loads of debate lately about whether or not they are going to be as much as delivering the following leap in AI. An organization known as Liquid AI is shaking issues up with its Liquid Basis Fashions (LFMs).
These aren’t your typical generative AIs. LFMs are particularly optimized to handle longer-context information, making them perfect for duties should deal with sequential information like textual content, audio, or video.
The LFMs obtain spectacular efficiency with a a lot smaller mannequin, much less reminiscence, and fewer compute.
NVIDIA opens up
Nvidia simply dropped a game-changer: an open-source AI mannequin that goes head-to-head with massive gamers like OpenAI and Google. Their new NVLM 1.0 lineup, led by the flagship 72B parameter NVLM-D-72B, shines in each imaginative and prescient and language duties whereas additionally leveling up text-only capabilities.
With open weights and NVIDIA’s promise to launch the code, it’s getting tougher to justify paying for proprietary fashions for lots of use circumstances.
Simply say know
A brand new examine discovered that the most recent massive language fashions (LLMs) are much less prone to admit once they don’t know a solution to a person’s query. When customers ask these fashions a query they’re extra prone to make one thing up reasonably than admit they don’t know the reply.
The examine highlights the necessity for a elementary shift within the design and improvement of general-purpose AI, particularly when it’s utilized in high-stakes areas. Researchers are nonetheless attempting to grasp why AI fashions are so eager to please us as a substitute of claiming: ‘Sorry, I don’t know the reply.’
AI inside
It looks like everyone seems to be slapping an “AI” label on their merchandise to tug in clients. Listed here are just a few AI-powered instruments which are truly value trying out.
Bluedot: Report, transcribe, and summarize your conferences with AI-generated notes and not using a bot.
Guidde: Guidde magically turns your workflows into step-by-step video guides, full with AI-generated voiceovers and pro-level visuals, all in just some clicks.
In different information…
Listed here are another clickworthy AI tales we loved this week:
And that’s a wrap.
When you dwell in California we’d like to understand how you are feeling about SB 1047 getting vetoed. Is it a missed alternative for AI security or a constructive step that can see us get AGI quickly? With highly effective open-source fashions like NVIDIA’s new bombshell, it’s going to be tougher to control LLMs anyway.
OpenAI’s Realtime API was the spotlight this week. Even should you’re not a developer, the prospect of interacting with smarter customer support bots that speak to you is fairly cool. Except you’re employed as a customer support agent and also you’d prefer to preserve your job that’s.
Tell us what you suppose, observe us on X, and ship us hyperlinks to chill AI stuff we might have missed.
