Sabitlenmiş Tweet
@lefttailguy funnily. the lead eng who built chrome are now at oai. sundar was the pm afaik?
English
sn1990u
320 posts
@a__tomala read, paint (acrylic is good and fun and cheap), pick up an instrument
English
on leave from work for an indefinite period of time
not really sure what i should do while away from my job..
English
Waiting for the inevitable surge pricing change and tbh I'm for it
Boris Cherny@bcherny
We doubled Claude usage on weekends, and outside 5–11am PT on weekdays for the next 2 weeks.
English
incredibly unlikely tho
- they directly compete on ads (unlike say apple where android is not directly revenue generating)
- meta poaching engineers doesnt help
- if meta managed already a 2.5 quality model theyll do fine in a year (this would be an acclerant only)
Eric Seufert@eric_seufert
If it is true that Meta licenses Gemini to power MetaAI, and I find it somewhat dubious, it would be an incredible coup for Google and give Gemini a credible claim as the world’s most scaled model by consumer use (~1BN iOS devices, 1BN MetaAI MAU as of last May).
English
@eric_seufert Ah US. You're right. Globally esp emerging markets dominate total (global) youtube watch time and those are very very mobile skewed (India indonesia Vietnam particularly)
English
Worth noting that the television is YouTube's principal device by watch time.
English
at this point after seen 1 in 4 pretraining runs fail they seem more like biological beasts than scientific experiments. shit that didn't show up in smaller ablations rears its head at scale, you 2x compute & bsz gets messed up. shit is hard to scale. not surprised
Ethan Mollick@emollick
Both xAI and Meta seem to be falling behind, based on the Grok 4.2 benchmarks and this reporting. Frontier AI models are really a three way race at this point.
English
what's gemini's usp? what makes it unique vs chatgpt? at least chatgpt came first and has amazing word of mouth
Bojan Tunguz@tunguz
My take: the reason why Gemini is not being used more, despite objectively being one of the most capable models, has a lot to do with Google's sh**y big corporate marketing strategy. They have no unhinged posters there.
English
yes. but ideally you wanna do both. ocr + scrape as much as possible and pass it to the context window
Nan Yu@thenanyu
Is it better to just render PDFs as images and OCR them rather than read directly then?
English
A lot of people seem to think Sam Altman's comments here are inherently demeaning to humans or something and I don't get it at all. youtube.com/watch?v=CujJ74…
YouTube
English
@Teknium @chongdashu html xml svg etc is amazing because their crawling/scraping is top notch imo
English
If there was one takeaway from the Gemini 3.1 Pro launch …
It’s that Google is all in on SVG
Junior García@jrgarciadev
I'm building a node-based tool that turns any SVG into animated SVGs using Gemini 3.1 Pro it preserves the original aesthetics and the results are insane
English
compaction works the same across models - when your session hits the contextTokens limit (200k default), openclaw summarizes older messages to free up space. with claude you can push to 1M (context1m: true + contextTokens: 1000000) and prompt caching means re-sending context is cheap. gemini handles big contexts worse so you want a lower cap to force compaction earlier
native tools = using the models built-in function calling format. separated agents = small focused agents per task instead of one mega-session. both keep context way smaller per session
English
For weeks OpenClaw on Gemini drove me crazy.
Contexts exploded to 400k+, sessions died mid-task, token bill quietly ate my soul.
Every “pro tip” made it worse.
So I finally opened the source code + Gemini docs and started over.
English
@MadGarizs noob here with no knowledge of openclaw so apologize for dumb qs. can you explain how compaction works in case of claude? same q for native tools, separated agents?
English
Real takeaway: you can’t just port a Claude config to Gemini.
Gemini needs its own rules - hard caps, proactive compaction, native tools, and separated agents. Everything else is wishful thinking.
English
sn1990u retweetledi
like if you’re even kinda good at kubernetes or C++ you’re unbelievably employable rn
English
every other ai researcher i talk to want to start a coffee shop / cafe after everything is over.
i was actually seriously looking into this. and then i realise we should be able to get AI (e.g., Gemini) to do this automatically for us. Lease a place, negotiate a good price, do branding, marketing, hire baristas.
they should be even be able to do "human-use" (and call humans as tools when they have to).
and then they should be able to be able to scale this up. run multiple businesses, cafes, pet shops, tuition centers, grocery marts etc...
English
guys read jax-ml.github.io/scaling-book/i… by
@_sholtodouglas @jacobaustin132 and few others its really quite good I promise you
Jared L Kubin@JaredKubin
The most important 40 seconds of the Dario pod…. Why do these models feel like magic all of a sudden (fintwit)!? “KV CACHE OPTIMIZATION”…which is still a major major bottleneck btw…
English
@a__tomala bay area def omega intense + commutes kill leisure time
not drinking / socializing as much (time + energy spent on fitness does bizarrely get in the way)
theres probably other social factors - angstiness/increased taboos around workplace romance
English
I wonder how much of this is due to people working very hard. After years of working long hours, I’ve gradually lost my ability to experience romantic attraction.
This has gotten worse as I’ve worked more intense jobs and it seems like many SF jobs are similar intensity now.
Ali Shobeiri@Ali_Shobeiri
Half of men in San Francisco reach age 40 without ever getting married. In New York City, it's over a third. Surprisingly, the problem is just as bad for women. Across every age group past 25, both NYC and SF lag behind the national average in marriage rate:
English
Apple Nvidia don't like each other.
Manz🌪@ManzTrades
Every once in awhile, I ask myself why is the 2nd most valuable company in the world, completely sitting out the largest capex investment cycle ever?
English