devcycle
1.2K posts
devcycle
@dev__cycle
looking for better software and bike rides
sf Присоединился Kasım 2023
243 Подписки48 Подписчики
Mythos is underhyped and overhyped at the same time
it's not even close to AGI
but coding and math is already well beyond median human performance and close to superhuman
which is why I also don't like the term AGI
it's a completely useless term
because when you achieve AGI, these systems will have aspects that are so far beyond superhuman it's not even funny
English
@teortaxesTex i think this is the best natural test i've seen yet of the "model degradation" phenomenon being psychological - people genuinely just get accustomed to new levels of intelligence *so fast*
English
Fable will be the most mythologized model this year probably
The first LLM we've had a chance to be awed by but not enough time to find the wrinkles
The hedonic treadmill is real, people get bored with miracles easily. Unless you instantly take them away
English
@EgeErdil2 and business value is based on profit, not flops. it's probably not a terrible proxy, but it's obvious that nvidia has better margins
English
@EgeErdil2 the long tail of random small "AI" applications run on CUDA only, but the non-OAI big bois (e.g. gemini/claude training/inference) are diversified to TPU/Trainium. I expect the big ones to 📈 and the long tail to 📉
I do not think the markets are correct here fwiw
English
NVDA datacenter revenue growth has been tracking my forecast from april 2025 that gave timelines to remote work automation on the order of 20 years
i had argued an exponential forecast was too bullish and a linear one too bearish, and projected the intermediate curve
English
@EgeErdil2 so you think gemini and claude are primarily inference on nvidia? i think all the evidence points to TPUs
English
@EgeErdil2 i refuse to believe that TPUs (which are training gemini and claude models) are a rounding error
English
@dev__cycle cerebras doesn't have enough scale to produce any relevant quantity of chips, which is being accurately priced by the markets
all compute producers other than nvidia are a rounding error right now
English
devcycle ретвитнул
@EgeErdil2 ah true. but i don't see any reason to extrapolate that to "nvidia will be the bottleneck to full automation" - all major labs are significantly on TPUs, Cerebras, maybe others. pretty sure OpenAI is designing its own chips as well.
English
@dev__cycle anthropic is using nvidia. they recently agreed w/ elon to use colossus 2, which is entirely nvidia chips
English
@EgeErdil2 errr, sure, i mean this is just one deal. i just meant to call attention to the fact that Anthropic is not using nvidia, so anything assuming nvidia is a majority is assuming anthropic isn't winning. not to mention that openai is moving to cerebras chips.
English
@dev__cycle "tens of billions of dollars" is less than 10% of NVDA's annualized revenue rn
English
i really truly can't wait
i just think we should give the defenders a head start
vixhaℓ@TheVixhal
One day, Mythos / GPT-5.5 Pro-level models will run locally on my laptop.
English
is anyone seriously working on killing off prop 13?
ryanrz.skr@Ryanrz
In Hollywood near multiple studios. 2nd largest U.S. city. Perfect climate. Strong economy. Housing shortage. And multiple vacant lots on the same block. I don’t want to hear shit about taxing trillionaires until we fix the tax and land use policies that enable this.
English
@celestepoasts no? wouldn't anyone who actually cares about this going well do that?
English
Sounds like the concern is China not jailbreaking like initially reported. I strongly suspect the outcome of this Anthropic/Fable 5 situation is going to be China gets blocked entirely from accessing any of the stronger Mythos models.
Axios@axios
Scoop: Anthropic flies staff to D.C. to clean up White House fight axios.com/2026/06/14/ant…
English