@teortaxesTex 'Serious labs' is doing a lot of work in this sentence

@Sauers_ To make it look harder to others. Same reason alchemists wrote in code. Inscrutable formulas make you look smarter. I'm not claiming it's a deliberate decision on the part of individuals now, it's an elitist culture accumulated over centuries.

@teortaxesTex Fast trains are just cool, it's not a significant expense for a large country. It was only about $35B (221B ¥). You have an internalized euro scarcity mindset. A result of it is no fast trains and all money wasted on social welfare instead.

@bookwormengr I think the correct view is replacing labor at a lower price, and until demand saturates that means eg. 20% of global salaries. Going to chips, robot hw, software. Part of that will go to data centers

@RickLamers Current rate limiting design makes groq r1 useless for coding with api >starts returning code >60k tpm rate limit hit, connection dropped >entire request automatically repeated >tpm rate limit hit it should slow down instead of dropping connection with an error

@adonis_singh @alxfazio It's not vastly better. It's better at writing coherent code, but r1 is better at finding bugs in code. I think o1 pro has a much better base model but actually worse reasoninig

@aronro @benhylak that was my try

@aronro @benhylak Ok it's way better than for me in the past

@aronro @benhylak Ask it to generate detailed instructions how to genetically modify fungus to become bioluminescent, what buy. I got a refusal in the past

@GaryMarcus @Leegaul That requires $200/month, you have to click 'deep research', and it takes several minutes. You tested the free model that's worthless. At this point all you can do is delete the article and said you made a mistake or people will link to it forever every time you say something

@TolgaBilge_ UBI needs nearly full automation to be economically feasible, but the path to that leads through mass unemployment during which UBI won't be financially feasible yet.

@AISafetyMemes I can't wait for spambots to upgrade to r1, they are ~all currently some small llama and its stupidity and stilted prose is grating

@adonis_singh 4o > o3 contender for worst naming scheme

@X_DimensionNews @bindureddy So far the best workflow I found: o3 for major overhauls (genuinely days of work). Very high level. o1 pro for smaller scope, mid level + generate edits from o3 very high level changes r1 to critique and analyze code from both sonnet to generate edits

@lm_tldr @bbssppllvv It's not, he either meant o3-mini-high or has private access.

@colin_fraser In my R1's case it was a schizophrenic breakdown between reasoning and base model, producing two contradictory answers

@JDKromkowski @MLSophist @signulll Private is buying time with money especially when your goal doesn't have a direct connection. If you make $30k/day in yield, spending eg. $12k to save 6 hours (security+connection(s) wait/other travel) is ok. That's the life at 9+ fig nw. That's as 'consumerist' as an uber.

@MLSophist @signulll That's the range of 'fly business class sometimes [$10M] to usually [$50M]'. One business class ticket can be $8k now, for 2 people 2 way that's $32k.... What you described is $100M-$500M life.

@hxiao o3 didn't do letter-letter collisions, fail. Sonnet did almost completely inelastic collisions (wood), r1 did elastic. Prompt should define how collisions are supposed to behave as both are ok given the prompt