Hacker News Reader

Comment by xyzzy123 | Hacker News Reader

Great, the start of model segmentation where I'm gonna need a legal license to ask about legal problems, a nutritionist license to create a meal plan, a medical license to ask about an x-ray, a pilots license to ask about a flight plan, be a registered electrician to ask how to wire something, etc etc. The licensing of allowed thoughts.

Apparently I can pay for partial solutions to the Riemann hypothesis but if my question involves a crackme or something that is an existential risk somehow.

View on HN

It's the open weight models that will save us from this fate.

OTOH, this is the cudgel that incumbents will use to get the government to protect them from open weight models.

I imagine the cudgel will be when the models are good enough and some moron sets their openclaw to hack a hospital or other critical infrastructure. I feel like we've learned nothing from the ransomware problems of recent history. Are we really saying it's a good idea to throw such capabilities in every emo teenager's hands because freedom?

The frontier labs would rather charge a premium for the best capabilities. You joke but they will probably soon have a GPT-bio that they license to Pfizer and a GPT-quant that they license to Citadel, rather than keeping those valuable capabilities in the public models.

> but they will probably soon have a GPT-bio that they license to Pfizer and a GPT-quant that they license to Citadel, rather than keeping those valuable capabilities in the public models

Nein. Neither Pfizer nor Citadel are going to use a hosted model - their whole business is IP. They will invest in self hosting models. Self hosting is becoming a big deal in the industries you're talking about.

Why?

Neither Pfizer nor Citadel wants to pay a supplier/vendor to build a core capability that the supplier/vendor will turn around and sell to their competition.

Citadel already is. They are happy to do it - bedrock is the host of choice. They are less happy about cost - but that is a question of value :)

Citadel are using which models on bedrock and for what purpose?

Core portfolio research? Update Confluence pages and JIRA tickets? Create marketing copy?

Sure, and OpenAI will be happy to let them self-host for a large fee. However I doubt that Pfizer will be developing their own frontier models for biological research.

Just today, 5.6-sol refused to elaborate on a security vulnerability it claimed it had identified in my code. It could plainly see the git history with my name on every commit going back to inception, but apparently elaborating on the nature of a purported vulnerability is too scary for openai.

It would fix it for me, automatically, if I was willing to let it run amok in my codebase. But asking if the "vuln" was exploitable triggered the guardrails and a "we can't show you this content" error because I am not part of the cybersecurty Trusted Access bullshit.

Could you ask it to write a failing test before fixing it?

I could also ask it to fix it and look at the diff, but that's rather beside the point, don't you think?

Oh totally, it’s pretty silly.

I had 5.6 refuse to back up my android tablet the other day because it involved root access.

Outstanding technology, truly amazing what man has managed to create.

I have heard AI can already fly (it carries your laptop to the ceiling when it does), but usually it refuses to demonstrate this capability for safety reasons.

I've heard it can ramp the fans up so hard it'll suck the gold right off your motherboard traces and send it to OpenAI. This is the thing that the US administration is worried about, because Chinese open weights models will copy the function and then send all that gold to China, and the US will have to dip into the stores at Fort Knox, and potentially devalue the greenback.

+1. And these ^^^ are exactly my thoughts for which I had been downvoted to oblivion before. TaDa. They materialise.

LOL ... and same people keep downvoting me. Right. Good luck.

Reply on news.ycombinator.com