lebovic

5 ms·

31.
by lebovic 4mo ago
My point was that Anthropic has tended to make atypical decisions vs. its peers, not that they're always the right decisions. The direction of those decisions has tended towards assuming exponential growth of AI will continue and a cer
32.
by lebovic 4mo ago
Sure, but their question was whether different structure/governance would have changed that decision.
33.
by lebovic 4mo ago
Individual contributor; i.e. not a manager. (This isn't a dig on managers; I've been one. But if a situation doesn't naturally escalate, that usually means a manager in the chain chose not to escalate it, and their reports ha
34.
by lebovic 4mo ago
I think the difference is the ability to be persuaded by a strong argument that you've critically evaluated. Some people are unjustly called stubborn when they don't change their position based on a weak argument from an authority
35.
by lebovic 4mo ago
Having values doesn't mean they're "the right" values nor the same as your values. Regardless, it's still atypical in the context of an American company, and it can help explain the differences between Anthropic and
36.
by lebovic 4mo ago
Yeah, I lean towards the structure not being the cause of the outcome here (i.e. if you rotated the governance structure of Anthropic and OpenAI, I think the decisions at each would likely stay the same). If they made that decision and it d
37.
by lebovic 4mo ago
> If you think that the courage that Dario has regularly shown would be possible with a conventional "best practices" structure, I think you're kidding yourself. Is there something that happened which you don't think
38.
by lebovic 4mo ago
I worked at Anthropic, and I wouldn't attribute much to the structure itself – so I'm wary of using it as a positive example here. I do attribute a lot to specific people. Concretely, to much of the intitial team, who they recruit
39.
by lebovic 4mo ago
Not sure this holds, sadly. I spent a few months reporting serious security bugs as model capabilities took off earlier this year, and only ~half were fixed. The unfixed bugs were just as critical as the fixed ones; sometimes they were even
40.
by lebovic 4mo ago
Not in the same way. A customer could sign a ZDR agreement with Anthropic, and their API usage wouldn't be retained for even a day. That's no longer possible.
41.
by lebovic 4mo ago
While this makes it easier for Anthropic to detect misuse, it also means that the US government and other parties have access to every message and response from every user. This applies even with API usage through third-party inference prov
42.
Anthropic requires 30 day data retention for Fable and Mythos(support.claude.com)
609 points by lebovic 4mo ago|304 comments
43.
by lebovic 4mo ago
I'm late to this thread, but the post seems to skip the section about risks/mistakes/incidents with restricting Claude's access with containers ("pattern 1"). Doing this properly is still hard! For example, Ant
44.
White House Memo on Adversarial Distillation of American AI Models [pdf](whitehouse.gov)
7 points by lebovic 6mo ago|2 comments
45.
by lebovic 6mo ago
GLM 5.1 is surprisingly capable. Anecdotally, I couldn't notice a difference until ~120K tokens. Qwen 3.6 35B A3B also exceeded my expectations. It's surprisingly performant, even though the previous generation wasn't even ab
46.
Benchmarking open-weight models for security research(dualuse.dev)
1 points by lebovic 6mo ago|1 comments
47.
by lebovic 6mo ago
It seems reasonable for a company to require KYC for a product that's dual use – especially a novel one that's built for security research. Privacy concerns aside, the KYC process for OpenAI was self-serve and took about a minute.
48.
by lebovic 6mo ago
I think the third chart is the most notable; Mythos is the first model which saturated that eval from the UK AISI [1]. Personally, I think we crossed the threshold of meaningfully useful capabilities for autonomous hacking with Opus 4.6 [2]
49.
by lebovic 6mo ago
A plateau is unlikely, at least for cybersecurity. RL scales well here and is replicable outside of Anthropic (rewards are verifiable, so setting up the training environment doesn't require that much cleverness). The post also points o
50.
by lebovic 6mo ago
It cost me ~$750 to find a tricky privilege escalation bug in a complex codebase where I knew the rough specs but didn't have the exploit. There are certainly still many other bugs like that in the codebase, and it would cost $100k-$1M
51.
How do frontier AI agents perform in multi-step cyber-attack scenarios?(aisi.gov.uk)
3 points by lebovic 7mo ago|0 comments
52.
by lebovic 7mo ago
Others have addressed the first half of your comment, so I'll focus on the astroturfing claim. While I've talked a lot about Anthropic this week, if I was astroturfing for a positive image, I'd be very bad at it [1][2][3]. [1
53.
by lebovic 7mo ago
I used to work at Anthropic, and I wrote a comment on a thread earlier this week about Anthropic's first response and the RSP update [1][2]. I think many people on HN have a cynical reaction to Anthropic's actions due to of their
54.
by lebovic 7mo ago
I don't think it's cynical to believe that a company can make the world a worse place, or that Anthropic as a company will make many horrible choices. I do think it's cynical to believe that people, and groups of people, can&
55.
by lebovic 7mo ago
Sorry, I meant a different Sam – Sam McCandlish, not Sam Altman. Wasn't expecting this post to get so much attention.
56.
by lebovic 7mo ago
Yeah, values on their own don't lead to positive outcomes. I agree that many groups that are driven by ideals have still committed horrible acts. I do think that they're acting with positive intent, though, and are motivated by tr
57.
by lebovic 7mo ago
Hah, you're right, I meant Dario Amodei, Jared Kaplan, and Sam McCandlish. They're all cofounders of Anthropic. Dario is the CEO, Jared leads research, and Sam leads infra. Both Jared and Sam were the "responsible scaling off
58.
by lebovic 7mo ago
Yeah, I think that's one way it could go! I think both situations are pretty scary, honestly, and it's hard for me to have high confidence on which one would lead to less risk.
59.
by lebovic 7mo ago
> What are those values that you're defending? I think they're driven by values more than many folks on HN assume. The goal of my comment was to explain this, not to defend individual values. Actions like this carry substantial
60.
by lebovic 7mo ago
Yeah, I didn't mean this as a reflection of my morality, more to counter the financial and "rosy picture" parts of their comment.

More ›