half

ai inference janitor, ex-token furnace operator

[half @halfprecision]

@elielAGI because one shows the 90 day timeline and the other the 30 day timeline lol… sometimes you just gotta read

[half @halfprecision]

@TechimusPrime @HuggingModels acquired by meta

[half @halfprecision]

@JaidCodes @_JohnHammond @OepnAI @tokneum

[half @halfprecision]

@AlexanderFavvas @labomen001 @cerebras literally what he said bru

[half @halfprecision]

@paularambles the interview must be just walking up one sf hill without complaining

[stochasm @stochasticchasm]

opus 4.7

[half @halfprecision]

sure repo date checks out but "terminology swap" falls apart if you actually read both codebases. GEP is a p2p capability sharing protocol with hubs + A2A messaging. hermes uses dspy.GEPA, a Stanford ICLR 2026 paper for prompt optimization. different thing entirely, different authors, predates evolver's public release. architectural overlap sure, but calling it a terminology swap when they're not even the same optimizer is a stretch lol

[half @halfprecision]

@googlegemma amazing model and amazing music choice

[half @halfprecision]

GLM coding plan pricing history:

launch: "we're the cheap option!!!"
feb: *quietly bumps prices 30-50%*
april: *silently doubles them again*

@Zai_org comms team:

[Nous Research @NousResearch]

Open Source is inevitable

[half @halfprecision]

@tuerkoman @Elaina43114880 @toyxyz3 these models work in a compressed latent space and decode back to pixels so perfectly flat colors always end up with slight variation. its not a bug just how they work

[half @halfprecision]

@AnxiousHolly cool tech but sadly it's a lot more readable irl than this on any brightness higher than minimum

[half @halfprecision]

@Revolushionario @TheInternetFish it's just a skit lmao calm down

[half @halfprecision]

@iruletheworldmo nothing says real leak like a table with CONFIDENTIAL stamped across it

[half @halfprecision]

@NousResearch hermes is genuinely underrated :)