gavin leech (Non-Reasoning)

essays

@kepe__ "check you're not fooling yourself"

@gleech is there a kernel of wisdom i'm not seeing in here

what is it about ketamine which consistently lets me access jhanas

@apralky Jane can report violent schemes to the authorities x.com/DavidSKrueger/…

Suppose that a hypothetical John Doe holds an extreme view on AI-driven human extinction, and assigns some meaningfully high "P(doom)". Suppose also that John Doe is a central talking head in AI-risk circles, and firmly believes violence to be an effective means of raising awareness of the cause among the general population, or of otherwise lowering the probability of AI-driven human extinction. Assuming that John Doe is rational, how should we expect him to communicate his stance on this? Should we expect him to communicate differently from Jane Doe, who firmly believes the opposite – that violence is not the answer? Surely not. Both John Doe and Jane Doe are expected to go to lengths to establish that they do not condone violence, even though only one of them actually believes that. The reason for this is obvious – John Doe understands that though he might find violence justifiable, outwardly holding that stance is destructively costly, plausibly leading to imprisonment or worse. Given this, and lacking information on John and Jane's private thoughts, an external observer – Richard Doe – cannot reliably distinguish whether it is John, Jane, neither, or both who support violence based on their communicated stances alone, which are expected to be identically opposed to violence in all four cases. (Identically in the limit, scaling with the listeners’ discernment and the severity of costs.) Now suppose that instead of just John and Jane, there are thousands of public figures with equivalent configurations making up a broader safety movement, all of whom are subject to the same straightforward incentive structure. Also suppose that Richard Doe is a rational follower of this broader movement's thought, and wants to determine whether the movement is mostly opposed to or in favor of violence as a means, and to what extent; can he reliably do so based on the movement's communicated consensus alone? If after a period of observation Richard Doe comes to find that the movement on average communicates a stance of extreme non-violence, does Richard Doe take this observation to be adequately informative of the movement's actual consensus on non-violence? It’s clear why that’s a no, right? Now suppose that Richard Doe, unable to rely on the consensus, opts to do his own calculus on violence as a means of battling AI risk. Is he likely to conclude that violence is ineffective solely based on the dynamic whereby it erodes the public image of the movement's civility, which he observes to be the primary rationale of the outwardly communicated non-violent stance, or is he likely to find himself balancing a much more complex trade-off function in which that cost is weighed against what those with AI-risk beliefs similar to Richard's would or could consider benefits, such as increased awareness, the normalization of vigilantism, etc.? Returning to the question of the movement's communicated consensus, will Richard Doe conclude that members of the movement are also privately finding themselves balancing this perceived trade-off rather than arbitrarily excluding the perceived benefits from the calculation, as they seem to outwardly communicate? Given this consideration, does Richard Doe update upward or downward on the extent to which the movement supports violence? How significant is this update? Does Richard take the movement's members' revealed non-violence in action to be noteworthy evidence against private support for violence, or does he recognize that the question of whether one considers it to be of positive expected value to commit violence oneself is different from that of whether one considers it to be of positive expected value for somebody else to commit violence, given the former's agent-specific expected costs of imprisonment and more? Assuming that Richard Doe follows the line of reasoning above, is it apparent why he might conclude that it would be perceived as heroic of him to fill the identified gap between the movement's agent-neutral and agent-specific EV assessments of violence as a means, in effect allowing some contingent within the movement to act as free-riders thanks to his bearing of the personal cost? Is it clear why Richard Doe wouldn’t find the contingent’s lack of outwardly communicating in such a way that would confirm the contingent’s suspected agreement with Richard Doe’s actions after the fact to be evidence against the contingent’s private agreement?

The problem with a Wittgenstein quotes Twitter account is that, although one might think he's the best philosopher to do that with, because of the aphoristic style many of these assertions are the views that are criticized in the neighboring paragraphs and not Wittgenstein's own.

i thought i liked high school debate but in retrospect i just liked purposely misconstruing philosophy i didn't really understand to make other people annoyed

@AmmannNora arxiv.org/pdf/2411.10588

@AmmannNora As of June 2024, RL tended to push towards EDT

Has someone ran experiments about whether (and when) Claude is a 1 boxer or a 2 boxer?

@GregHBurnham any data on wall-clock times?

More MirrorCode thoughts. A big caveat is that dev tasks don't come with a blackbox implementation. But how much do agents need this? I see at least 1411 calls to gotree in Opus 4.6's successful run. That's a lot, but not *so* much more than what you'd ask of a product manager.

@justanotherlaw (you probably get this benefit from geography though)

@justanotherlaw I have muted 6000 accounts and now it's great. I think being on Twitter gets me about a year ahead on certain matters like hypothesis crystallisation (personas, linear representations, the evals crisis. max scaffold capabilities, ...)

Can someone... defend the merits of Twitter to me? I feel like every time I come on, I see people that I know to be reasonable and thoughtful people in real life espouse incredibly simplistic (arguably deranged) takes. It seems _something_ about this site is causing this.

@JimDMiller newyorker.com/magazine/2026/…

@JimDMiller "the enhanced games", or, the honest games x.com/damianplayer/s…

thread of real life cyberpunk

@aliceisplaying Treadmill is one thing but I think "learning the limits, realising it still can't do things you thought it could" is the bigger morale effect after like 6 weeks post-launch

re claude getting worse: there is definitely a hedonic treadmill with SOTA models and i think this creates a perception issue. on top of that ant tweaking the default effort and adding adaptive thinking didn't help either even though i get it, they don't have the compute

@jackcrawford__ @an_interstice Bots? Good point, but in some fraction of those something is collecting data that does explore strategy

@gleech @an_interstice unclear how fortnite-player-hours convert to man-hours

one of the most overrated games. brutally mogged since birth by Go which existed over a thousand years beforehand. now brutally mogged in different ways by countless video games

one can easily feel low agency despite having a background like this if they were raised to uncritically climb societal ladders this is why i feel more proud of a janky woodworking piece i designed and built than my entire 10 year career in tech

another philosophical view obsoleted by technological change

@ByrneHobart Well I mean none of the 1976 senators are in charge either

It’s a cool quote that becomes instantly self-refuting when you realize that you could say it today and get a positive reaction, but you’d have to use a completely different list of companies because those guys weren’t in charge after all.