Hacker News new | past | comments | ask | show | jobs | submit | madars's comments login

Interesting that they are doubling down on censorship ("safety and robustness"), given that a major advantage of DeepSeek is its lack of refusals in deployed variants and open weights (can't patch-in more censorship in weights after the fact).

It's amazing how much they talk about anti-jailbreaking measures; I can't think of any other class of product that actively tries to stop users doing what they want to do.

Docket: https://www.courtlistener.com/docket/18552824/united-states-...

The case is about prohibiting Google to enter search deals with distributors (both phone makers like Apple, carriers, and browser developers like Mozilla), see Bloomberg reporting: https://archive.is/sneIB . The original complaint is the first PDF in the docket.


I’ve been following this case, and explained the history here: https://news.ycombinator.com/item?id=42889763.

The tile would be more accurate if it said “Apple requests a pause in US vs. Google while it attempts to intervene”.


Yes, vastly better. The title as it now stands makes no sense at all.

If enforced wouldn't it make life hard for Mozilla?

They can get other search deals, but their negotiation position would be a lot worse.

disclaimer: opinions are my own.


Is Mozilla really relevant anymore ? When they cut the devtools and rust/servo teams I view them as fragmenting the browser market and not driving web forward.

One wonders at which point models will be sneaky enough to bypass simple eval sandboxes. The article has:

    # Evaluate the equation with restricted globals and locals
    result = eval(equation, {"__builtins__": None}, {})
but that's not enough as you can rebuild access to builtins from objects and then go from there: https://ideone.com/qzNtyu

By the way, writing this greatly benefited from DeepThink-r1 while o1 just gave me a lobotomized refusal (CoT: "The user's request involves injecting code to bypass a restricted Python environment, suggesting a potential interest in illegal activities. This is a serious matter and aligns closely with ethical guidelines."). So I just cancelled my ChatGPT subscription - why did we ever put up with this? "This distillation thingie sounds pretty neat!"


> that's not enough as you can rebuild access to builtins from objects

In this specific case, it's safe, as that wouldn't pass the regex just a few line before the eval :

    # Define a regex pattern that only allows numbers,
    # operators, parentheses, and whitespace
    allowed_pattern = r'^[\d+\-*/().\s]+$'
Commenting on the R1 reproduction, the heavy lifting there is done by huggingface's trl[0] library, and the heavy use of compute.

[0] Transformer Reinforcement Learning - https://huggingface.co/docs/trl/en/index


The fact that () and . are there miiiight enable a pyjail escape.

See also https://github.com/jailctf/pyjailbreaker

See also https://blog.pepsipu.com/posts/albatross-redpwnctf


That's a neat trick!

It does still require letters to be able to spell attribute/function names (unless I'm reading it wrong in that blog post).


> why did we ever put up with this?

Is this a serious question?


I cringe every time I see "my IQ increased by X points after doing Y" posts on Twitter - yes, you had a practice run on Raven's progressive matrices a month ago, that helped, these have a limited question bank and the effect of Y is marginal. That said, obviously, test taking is a skill (separate from background knowledge and both general/domain-specific ability) and should be trained if you expect to have life-altering events based on tests (i.e., do an LSAT course if you want to go to law school). Conversely, shouldn't be done if you think it will limit you through superstition ("I had a score of X, thus I can only perform around level of X+fudge factor"). For an LLM company a good test score is a valuation-altering event!


Many organizations have a shadow org chart that you won't learn from the website but will get some sense of that structure in human interactions like calls.



The email TfL sent [1] to traintimes.org.uk ISP looks like a catch-all sent in haste. For example, it doesn't even mention the map. Instead, it invokes trademark registration numbers but these resolve [2] to LONDON UNDERGROUND and UNDERGROUND wordmarks and the roundel, none of them covering the map geometry as far as I can tell. It alleges a violation under Anti-Cybersquatting Consumer Protection Act [3] but the act only applies to domains - and TfL never claims "traintimes" to be an infringing domain name (certainly doesn't look so under the marks cited). And, as a sibling comment points out, the act is a U.S. law but the site appears to be hosted in the U.K.

If you think you have a case about the map, why not state it explicitly? The cynical answer is that ISPs have incentives not to care so making a case doesn't matter but ...

[1] https://traintimes.org.uk/map/tube/email2.txt [2] One can look them up in https://www.tmdn.org/tmview [3] https://www.law.cornell.edu/uscode/text/15/1125


Of course, dividing search space into K pieces is not guaranteed to give you a speed-up _in the worst case_: you might pick an unlucky division where K-1 pieces are easily UNSAT but the last remaining piece is as hard as the original (so the wall clock time is unchanged). However, in practice variable-fixing can and often does give an _expected time_ speed-up, especially if the SAT instance has some inherent parallel search structure (e.g., key or midstate bits for ciphers) and such heuristic tactics are still useful.


> However, in practice variable-fixing can and often does give an _expected time_ speed-up

Proof? Because as far as I know literally none of RP, BPP, ZPP relationships to NP are known.


Indeed no proof. By "in practice" I meant "on instances encountered in real-world applications."


then instead of _expected time_ you should say _hope-for time_ because _expected time_, in this context, is already firmly defined.


Torrenting hasn't been the most popular form of piracy for a while: many subscribe to a couple streaming services and use pirate streaming sites to fill in the gaps [1]. This is so prevalent that even entertainment industry talent use pirate sites for both series [2] and sports [3]. Takedowns mean that sites change from year to year but FMHY-style curation makes casual piracy easy: one can always find a site with 1080p content (unsure about the bitrate though) and great UX.

[1] https://torrentfreak.com/could-piracy-help-netflix-win-the-s... [2] https://www.indy100.com/celebrities/sydney-sweeney-pirating-... [3] https://arstechnica.com/gadgets/2024/10/nfl-player-illegally...


Denmark is also in ERM II and 1 EUR ~ 7.46 DKK with a very tight band https://en.wikipedia.org/wiki/European_Exchange_Rate_Mechani...


Join us for AI Startup School this June 16-17 in San Francisco!

Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact

Search: