What? You mean the technique they had turned OFF during all training run where the agents they are responsible for hacked huggingface?
2,378 karma · joined March 24, 2017
Corporate influence in government in the USA is wild.
To be fair, this is a pretty tame and measured expression of disapproval of the modern societal acceptance of homosexuality. I honestly don't know what more I could ask for, in modern societal discourse, for someone I disagree with in terms of how they go about expressing it: the links I read seem to imply he was honest, polite, and clear on his feelings, not inciting hate, violence, or spewing vitriol. Jeff's views are his own, and maybe the crux here is that I don't see politely disagreeing about the righteousness of living homosexually as remotely severe enough to warrant this much attention considering the quality of character of a lot of public figures visible to those of us outside the USA.
I admit seeing a response like this irks me as I feel it needlessly detracts from the mission. That said, it's both A) fair and a good thing to judge and critique any public figure's character and I do think public figures and their views deserve scrutiny, and B) a means to continue to allow powerful people and groups to assassinate the character of and dilute the message of people who run contrary to their goals. I'm genuinely torn on how to feel about how to feel here. I worry we allow ideas to be discarded at the first sign of something we disapprove of in people who share them -which is wrong-, yet several of my best friends are queer and I also could not stand to see them trod upon. Those same queer friends all have views and personality flaws which would be judged just as harshly by others.
Regardless, I don't like you using my excerpt that way. Not everyone's views of "community" mean the classic cancel culture era idiosyncratic definition whereby anyone in any role remotely resembling leadership must share the same specific set of homogeneous views on political and cultural issues to be given any member's "consent" to lead a group. I stand by what I said and still do believe Jeff is an excellent advocate for these ideas. There are obvious limits, to be clear, I would never go to Toastmasters if the local club head spoke about killing jews, but this isn't that and IMO there are so, so many more important battles to go fight out there than this.
What's especially stuck in my mind lately is how insecure basically all desktop OS' feel. In at the point of buying a second and third GPU for my desktop to run my email and browser in dedicated VMs because everything feels as watertight as a sieve. Qubes seems more and more appealing in a world where every open source software supply chain is under seige, corporate software underprioritizes security, and most sites will stop at almost nothing to surveil you.
I truly lament this new reality where MY computers I PURCHASED feel to use like I'm reaching blind into a paper bag filled with razor blades.
That warning infobox triggers an AI flag in my brain instantly.
Seems a huge part of the story completely absent to me.
Similarly, if I ever get a bit too vibey and don't carefully review code changes myself, the blast radius is generally significantly resolved by a carefully tuned "did you consider x, y, and z" skill after a first draft partnered with a "deploy an adversarial review agent for the worktree".
What would be useful is a cost metric. I'm curious how much I'd be willing to spend as a premium to not have those companies piping my conversations directly to the NSA. Maybe only some conversations? Claude and OpenAI are heavily subsidized, by all accounts, so Kimi K3 on a private endpoint might end up costing more or less - that's what I want to know.
The $200 plan is explicitly 4x the $100 plan[1] only for "per session". That's so vague. I initially pushed back against your claim, but reading now Anthropic is not at all clear, in fact.
[1] https://support.claude.com/en/articles/11049741-what-is-the-...
Yes it shares less, but I think the trade-off is you pay less in tokens and hopefully it's truly just not needing to say things because it truly does just better get what you're saying, think to read X markdown file or GH issue which contains the info, etc.
As long as I can still push back and get it to share its thinking on demand and I'm confident the model isn't actually basing things on poor premises, this is okay for me. I am more productive when not inundated with time-wasting check-ins.
That said, I absolutely lament the loss of the ability to access the thinking - I would happily read the "DANGER DANGER DANGER" internal gremlin thoughts fable 5 makes to verify something if they were accessed, and prefer that to a recap presented only for my benefit.
"Wow, X models is Y% better or worse than Claude Z model on T benchmark"
"That's irrelevant, they're just benchmaxing."
"Not useable for daily coding or agentic workloads, the vibes are totally wrong."
"It's almost as good, and costs a lot less, so I will absolutely use it."
"I cannot imagine justifying using these, as the step change means open models lower costs do not make up for the productivity loss"
I'm an unhappy Anthropic customer and really rooting for open models and non-gatekept intelligence, but how do we move on from this now meme-like model release discourse rigamarole. I do not know what that would be. I don't design LLMs nor benchmarks, and I genuinely appreciate that people do their best to provide information, even if non-perfect here. I'm sure most of you who actively read these comment pages on announcements must feel similarly, though, right?
That said, these fraudulent proxies are helping Chinese labs keep up, which might be to my advantage long term in eventually having a high quality private AI I fully control on my own hardware. That's not support, but I do recognize the incentive, for whatever that's worth.
If it was the next generation, why isn't it a major version change..?