qrdl

The first career that should be made obsolete is the AI programmer.

[qrdl @QRDL]

@VitalikButerin @shayne_coplan @mansourtarek_ @giancarloMKTS @aosipovich @Polymarket @metaculus @kalshi

I'm a huge believer in Prediction Markets. It is free speech, but with accountability. Messengers from the future, warning us of the consequence of our actions. But the markets have significant negative social utility when they are subjective and opaque.

It can not and it must not be just about gambling.

Prediction Markets have perverse incentives to spread disinfo and manipulate outcomes, so more work needs to be done to surface authentic signal and increase transparency to stop bad faith actors.

Polymarket, metaculus and kalshi are the leaders in this industry.

Polymarket is arguably the leader and best in terms of transparency and free speech, but has serious problems with bad actors and poorly written subjective markets.

Metaculus lack of trading means it doesn't have bad actors, and the rulesets are very well done, but their poor transparency is devastating and destroys a great deal of potential value. They also have poor accuracy and no emergent signal / breaking news because of their minimal incentives.

Kalshi has reasonable rulesets, but too many bad actors which is a result of their weak transparency. Through no fault of their own, DCM status suppresses their ability to embrace free speech.

I am hopeful, and perhaps even a little optimistic, that 2025 will be the year when Prediction Markets grow up. They will realize their greatest strength is their openness, that they have a profound responsibility to make the world better, and they are not here to just facilitate the transfer of wealth from the naïve and gullible to the sharp and cunning.

[Daniel Litt @littmath]

My general view has always been that it's not worth rushing to be the first to do work that someone else could do next week; clearly the marginal impact of such work is low. Worth thinking about this in the rush to publish results where the intellectual labor was done by AI.

[qrdl @QRDL]

@zerohedge I think he meant to say if 1000% of America's 1.5M developers.

[qrdl @QRDL]

@nasqret Memory is very powerful, but we need transparent and fine-tuned control over it. I am perplexed why the labs have not figured this out yet.

[qrdl @QRDL]

@d29756183 @AndrewCurran_ @Sean_Sooch18 True emergent would be something that appears in an unexpected, almost orthogonal manner - not billions of gpu hours of RLVR in that domain.

[qrdl @QRDL]

@d29756183 @AndrewCurran_ @Sean_Sooch18 I didn't see a scientific denial of RL, just a vague hand wavy "not explicitly". I mean, for real.

[qrdl @QRDL]

@KRavisrinivas @AndrewCurran_ @Sean_Sooch18 reinforcement learning verifiable rewards (RLVR) allows the AI to 'learn' outside the material we give it. It can try things, experiment, and get immediate feedback on coding and math. It can't do this for bio, so it just knows what they give it.

[qrdl @QRDL]

@joseph_h_garvin @AndrewCurran_ @Sean_Sooch18 Yeah, I'm exceedingly skeptical. "We spent billions of dollars on training and we accidentally made a massive cybersecurity profit center". Riiiiiiiiight

[qrdl @QRDL]

@AndrewCurran_ @Sean_Sooch18 Yeah, I think we'd need visibility into what's in pretraining. What does "patching vulnerabilities" mean? Anyways, yeah, RLVR is why these things are super intelligent for sure. Until we get that in other domains, we likely won't see the same 'emergent' capabilities.

[qrdl @QRDL]

@AndrewCurran_ @Sean_Sooch18 I'd like to see a cite saying them denying they did RL on cybersec. But even if that was the case, RLVR on coding/math likely enabled it. RLVR is not practically feasible on bio

[qrdl @QRDL]

@AndrewCurran_ @Sean_Sooch18 "Emergent" .. puh lease! They are training their models to be specifically capable in this area.

[qrdl @QRDL]

@SebastienBubeck more interesting than disproving it separately would be showing what can be proved/disproved based on the disproof.. This is a far more interesting capability. autonomously proving something is not as interesting as people think it is. https://t.co/diJjj51SPf

[qrdl @QRDL]

@wtgowers labs need to focus more on finding problems that are 'solvable', rather than just solving problems https://t.co/diJjj51SPf

[qrdl @QRDL]

@ankit2119 @__alpoge__ @_sholtodouglas Even assuming it was a neutral prompt (I'm skeptical) the point really is that finding which problems can be solved is half the struggle.

[qrdl @QRDL]

@wtgowers @Konrad_LV @prz_chojecki The declaration was pretty good and trying to walk a pragmatic line of bringing mathematicians into the new world. It's a tough spot to be in and I think it did a decent job of threading the needle.

[qrdl @QRDL]

@thomasfbloom Math is a tool, not an art form. It's an important tool and should be treated as such.

[qrdl @QRDL]

@formaliser_dude @prz_chojecki Formalization solves the problems. Formalized flooding can be dealt with at scale in an automated and effective fashion.

[qrdl @QRDL]

@jbailey @ben_j_todd @notanastronomer Lol, buddy: "In most jobs, you might not think of yourself as in competition against anyone at all"

[qrdl @QRDL]

@phone3301 @AcerFur Absolutely this, we need new programming languages that are AI native.

[qrdl @QRDL]

@prz_chojecki Interesting to see how reliable gpt5.5 is at informal reasoning. Must be trained extensively with lean in the background