On the "Why now", Dylan Patel from SemiAnalysis, the most prestige American Semiconductor industry research and analysis firm (who broke the Google DeepAI we have no moat neither does OpenAI in 2023) has the direct answer:
It’s Likely Worse Than We Thought
So, if we had to say “why now,” it’s likely to remove opposition to an extremely hawkish China export control bill:
after a large shift in Congress and heading into a lame duck session, it’s best to get your opposition out of the way to
get the bill passed. That is our best conspiracy theory take on the situation, and it lines up with the ripples from
Washington. So now let’s return to our original question.
100% on point for the bigger issue hiding in plain sight.
Nvidia employees actually like their boss regardless his personal life / (un)professional affairs, and petitioning the board to reconsider and generated 1000s of comments internally [1]. Like for two reasons,
- After all, Nvidia was one the BIGGEST if not the biggest millionaire minting factory for the computing era. And Jensen was the solo man behind this whole journey (he bragged about being the longest serving CEOs in Corporate America, too sad he jinxed himself now.) Now the millionare minting factory part is about to be superseded by Anthropic.
- His working style was well known. He's a huge supporter of Open Weight and invested the entire AI ecosystems.
Not just the optics, but the liabilities and class action lawsuits if they are not acting on it. After all, A board's fiduciary duty is the legal and ethical obligation of board members to act in the best interests of the organization and its stakeholders.
For the big corporations (consumer facing), they are not necessarily able to adopt change because:
1) they are under tremendous amount compliance boundaries, liabilities ambiguities and customer support issues, when agents go rogue or caused monetary damages to both service providers or merchants and consumers.
2) we collectively haven't figured out a way how agenic commerce would work. They have been some ideas floating around from Google, Shopify Cloudflare, but it nowhere near settled as a consensus.
3) Fraud/spam/cyber - Agents AMPLIFY these malicious interactions with their existing Human User Interface (HCI). In other words, the current HCI is not designed for agentic interactions.
For example: If your Meta Muse agent buys a $5k business class ticket from US to Europe using your Chase credit card that you find out after the fact, and you call your credit card companies for non-authorized transaction chargeback, who's responsible to the damages? Did you authorize the agent to buy tickets not?
If Chase denies your chargeback and make you liable for $5k to be paid, you pay for it or no? If you don't pay for it, United sues you in court, do you tell the judge you didn't authorize your agent to buy that ticket? Or you sue Meta instead? You see where this is going??
You have to be responsible for AI agent usage. Any other model doesn't work because then you can always argue "I didn't do it, my agent did" for anything.
If I say "I'm thinking of visiting China sometime" and the Meta agent proactively decides to extracts my credit card number from my Meta subscription information, and proceeds to buy a ticket without my knowledge... I feel like at some point I get to blame Meta? Clearly I did not give any sort of consent there.
Conversely, if there's a chat log of me saying "Buy a ticket to China" and then clicking "yes" on an approval prompt showing the full price, ticket dates, etc. then yeah, obviously that one is on me.
How do you make the agent's capability to spend money conditional on the A2H approval? Is it just an instruction in the prompt, or is there an enforcement mechanism? It seems to me like it would require cooperation from your bank (who needs to block transactions for which no A2H token is presented) and merchant (who needs to accept A2H tokens and communicate them to your bank with the other transaction details).
The reason HF didn't win a legal battle is because they didn't sue, and they didn't sue because OpenAI is their business partner (or at least are closely wedded through their mutual partner Nvidia). If HF really wanted to press charges, it's very likely they would have won on any demonstrable claims of damage.
If HF did have any hard feelings about the event, they were likely settled with OpenAI out-of-court.
I agree with the main point. Just a clarification that if Chase denies your chargeback request, then the airline gets paid, and if you refuse to pay your credit bill, it'll be Chase coming after you (via lawsuit or whatever).
Seems pretty clear to me that the liability for the unauthorized charge is yours, and that it’s between you and the credit card company until the volume of errors is so large the other companies need to take note.
I tried to register a new account with Google yesterday to use their latest model, but they banned me after, like, five minutes. They asked me to scan a QR code with my phone and even asked for a selfie. After I completed all the steps, they told me I might be using a bot and that my account had been deleted...
But apparently, it’s okay for that same Google to unleash bots to scrape entire websites...
This is very much inline with Greg Kroah-Hartman the long-time Linux kernel maintainer's view Anthronpic's Mythos audacious claims its cyber capability. Majority of those "findings" were way out of proportion and merely "pattern matching" previously discovered methods of exploitation.
Pattern matching is a large part of the work. That's why LLM's have been so effective. A human can't match the speed of the loop, which is simple but very labor intensive.
Apparently, while Gaussian splatting is not the core technique for all world models, it has become a dominant representation and rendering backend for spatial and reconstructive world models.
Most of the newer EV brands (at least in North America, e.g. Tesla, Rivian, Lucid) are very sensitive to automotive telemetry security. AFAIK, they are built by software engineers in house and operating like a software company than a traditional automaker whose software are NOT made by themselves but by 3rd party OEM auto suppliers who may or may not have the security mind when they release their product. In general automotive industry did not have the foresight in security engineering and neither did they do that when it was heavily mechanical and local systems instead of IoTs and electronics when they started.
So the author exaggerates their claim of their work can affect any car in United States.
Every telsa API calls are using the latest TLS 1.3 with X.509 for key card access. Their software updates VERY FREQUENTLY. And they don't use the *-telemetry.com domain that was demonstrated in their hack.
I came to check and say the same as well. I was thinking if they used AI to convert Zig to C.
And that is with me and everyone else knew it was coded in Zig in the first place. Had anyone not knew what Ghostty were I seriously think they would have thought it was a "C Library".
In the Github issue filed, the context[1] for session start was authored by a Postman engineer [2] whose Github profile[3] page tagline is "I Solemnly Swear That I Am Up to No Good."
Yeah, I am sure that engineer was trying to be witty and clever, but now if Postman's ever got sued in court, the joke's on them.
reply