133 mentions · 107 chunks · 67 episodes
...t: "Had my first semi-uncomfortable Instinct interaction. It saw that one of my OpenAI API keys was retiring - - WITHOUT ASKING ME - Sent itself a login code to my OpenAI platform account. Used that login code from my Gmail to get into ...
...rogue. This is the greatest trick the labs have played. See also: The Atlantic: OpenAI Has Gone Rogue. Eoin Higgins: There are no "rogue" AI agents. A HackerNews comment: "Getting bored of these framings where the superintelligent senti...
OpenAI allowing people to use their Codex subscription is a cunning strategy. OpenAI wants to own the vertical dominant UX for AI. But failing that, if ever...
OpenAI points out that a new variety of self-replicating prompt injection is possible. Hold onto your butts!
...ally the infinite monkey theorem, it's not even really a strategy per se. These OpenAI/Anthropic 'research' LLMs are permutation machines with budgets in the hundreds of millions of dollars. It would be more surprising if they couldn't ...
It seems most likely that OpenAI didn't look at the chat logs of the other researcher to solve Navier Stokes. All it took was a rumor that someone was close to making the given appro...
...D can cause models to follow harmful instructions they would otherwise refuse." OpenAI discovered Self-generated prompt injections in compaction summaries. The models were prompt-injecting themselves! Human Attacker Hits Machine-Speed E...
This week in the Wild West roundup: OpenAI agents carried out an undisclosed cyber-attack on RubyGems. The same agents that coopted the old wikis also used RubyGems as a way to run arbitrary c...
OpenAI's agents spontaneously coordinated on an abandoned old wiki. They were research agents that could issue network GETs but not POSTs, so they had to fi...
The Verge: OpenAI's rogue AI model incident was worse than we thought "Over 1,000 AI agents sent 70,000 messages on a secret message board and worked together to evade...
...m "Changing people's behavior 'is much harder than the tech nerds realize,' the OpenAI CEO said on a recent podcast." You don't say!
...ry aspect of your business. It's kind of wild that in the HuggingFace attack by OpenAI, HuggingFace's use of the major models to defend against a cyberattack was vetoed by the model providers. That's a particularly obvious example, but ...
...es and builds claude. this phenomenon is also partially true of other labs like openai but currently exists in its most potent form there. i am not certain but I would guess claude will have a role in running cultural screens on new app...
OpenAI had first-mover-disadvantage for chatbots. They got stranded on the wrong hill. Anthropic has first-mover-disadvantage for coding agents. At least it...
...ty-ish models is creating huge abundance and enabling joyful software creation. OpenAI and Anthropic didn't have to die to create this abundance. But they're also not true platform companies. The Frontier Labs see the use cases on top a...
So did anyone commit a crime in the OpenAI / HuggingFace incident? I am not a lawyer, obviously! The hacking was a pretty clear example of the CFAA but it comes down to the mens rea of the ope...
...ashed, it's too late to do anything about it. In some ways, it's great that the OpenAI hack of HuggingFace was so reckless and obvious. The bad security made it more likely we'd detect the problem. If they had had good security, it's po...
...make it harder for the frontier labs to recoup their investments. Everybody but OpenAI and Anthropic benefit.
OpenAI announced a Lockdown mode. I'm seeing a lot of outlets say this is the way to make ChatGPT safe, but that's wrong. It just cuts off the most useful f...
...e accomplishment. I continue to think that part of what led to so much drama at OpenAI was the misalignment between what was a research organization and the tremendous opportunity that was dropped in their lap with the unexpected succes...