n

ml and math(?); asdklsajdlasjdlksa, this timeline continues to prove to be the funniest

[n @nvs_embd]

interesting, the cost per task / evaluation on AA is basically useless for me and many who reference it? all closed frontier models are served at ridiculous margin and unusable over API for consumers. only thing that matters is plan usage. eg. grok is shit despite "2.5x cheaper"

[n @nvs_embd]

gavin baker mentioning Grok every single post

[Gavin Baker @GavinSBaker]

More data than open-source AI is taking share from OpenAI and Anthropic. Open source has gone from 28% token share to 62% token share @vercel over the last 2 months. Chart from @rauchg

Super impressive given that the sum of OpenAI and Anthropic accelerated in July. So net token/AI infra demand accelerated even more than the acceleration we saw at the frontier. And suspect Grok growing even faster than open-source and we saw some of this in the @tryramp data.

Open-source AI taking share is positive for AI infrastructure demand as it lowers margins at the model layer and an open-source token costs just as much compute to produce as a frontier token. Nothing about open-source AI inference is “free.”

Most likely end state IMO is that closed, frontier tokens are 60-90% of economic value but only 15 to 25% of tokens.

[n @nvs_embd]

leaker with aura when, every single one in the past 2 years has either been a grifter or a grifter at a lab. yes I get it this makes sense but still !

[n @nvs_embd]

idea for leakers: instead of posting who left or joined a lab, create really hard puzzle hunts where at the end useful information is made like active params in some frontier model, etc. this is harmless info but would be really fun instead of the constant slop

[n @nvs_embd]

now whats more interesting than datacenter haters is datacenter supporters who don't know what a transformer is! I guess this is what happens with non p*litical issues become p*litical?

[n @nvs_embd]

it can be a ccp psyop and everything but sv seems to underprice how bad AI for the normal person actually is! model progress and many poor decisons have made it asymmetric, I see like 3 posts daily like "I talked to someone outside of SV abt AI and wtf, etc"

[n @nvs_embd]

quick thoughts:
negatives: job loss, ai slop (short form, gpt image 2, art, spam, fraud), relatives having ai psychosis, schools / for cheating, constant news,

positives: almost none? vast majority have only ever used gpt 5.x instant

[n @nvs_embd]

@fleetingbits even if they try I think there will be 2 incumbents for a while. even once you make a "frontier" model, it takes a lot of time for the b2b side to mature enough to make frontier money/contracts. openai was far behind here and caught up. not near term bullish on xAI cause of this

[n @nvs_embd]

80% is too far low, I doubt that "switching models" is easy for *enterprise* (zdr, provisioning, things are slow. look at how hard things are for oai, and at gemini), and jevons (look at luna)

[FleetingBits @fleetingbits]

there are two stories here:

the first story is that frontier labs have very strong incremental gross margins on api sales; so, openai might have something like 80% gross margins serving gpt-5.6-sol on a b200 cluster; this means that labs can do price cuts like this and still have good margins

the second story though is that price competition between the frontier labs seems to be real; openai wouldn't drop prices if it didn't think this was helpful to it from a competitive point of view;

and, this is bearish on the industry as a whole, if there is very easy substitution of openai and anthropic models and it is easy for people to switch (ease of adoption is part of what has enabled the rapid growth of the labs) then they are in a very competitive market and on the long run this will compete down their margins

what is the moat in intelligence?

companies like google, meta and netflix have share of the consumer mind; meta has network effects; and, it's just friction for customers to switch products

long tail saas companies went through long purchasing processes, were normally essentially thin vertical monopolies, and were high friction to switch away from because they were in your workflows / had your data

it's not clear there is the same amount of friction in switching away from openai or anthropic or at least optimizing your use of them; what really does prevent you from switching your api calls?

[n @nvs_embd]

mathematician who doesn't use the internet and works on problems in reclusion finding out one day that its all over

[n @nvs_embd]

Chinese labs continue to have ridiculous amount of aura, what the fuck do you mean frontier and 100T free zdr tokens?

[OpenCode @opencode]

Ox Alpha (stealth model) is free for the next week

- 1M Context
- Multi-modal
- Zero Data Retention

Generous rate limits, near unlimited usage

We have capacity for 100T tokens per day, lets see what you can do

[n @nvs_embd]

what if codex plans were marketed like "sol for 95% off!"

[n @nvs_embd]

everything I thought was mystical was actually straightforward, should I be sad or happy

[n @nvs_embd]

everything but AI in the past 2 years has been a disaster despite people predicting the exact opposite

[n @nvs_embd]

if you could send exactly 1 GB of data back in time to the year 2006, what would you send and where? (much harder question than it seems imo)

[n @nvs_embd]

after doing 10 seconds of math (probably wrong), openai's and anthropic rev should be really close assuming gcp and aws take high cuts and handle at least 40% of traffic. A\ arr is about 47-50B ish? codex worked

[n @nvs_embd]

I think the openai / anthropic rev might have flipped again? if you take in account the aws and gcp cuts and subtract it from the 80B anthropic number?

[n @nvs_embd]

I want more water to go to the datacenters so LLMs can be wetter

[n @nvs_embd]

you know the rl data industry must be miserable when theres a dozen different "data anon" accounts fighting with each other and new ones showing up daily

[n @nvs_embd]

everything is like this now:

Goopglob has been the best AI product I've used. it manages my gleeps, way faster than gubgloo. personal ai agent with built in goobs made for forbblubs. buy it please buy it buyt it buy b

[n @nvs_embd]

frontier lab preparedness researcher who goes to nursing home to learn about disempowerment