I'm #124 of 1,000 initial contributors building VibeBench, the coding benchmark that measures what actually matters — how models feel when used on real work by experienced software engineers. vibebench.standardagents.ai/?id=galer7
English
Gabriel
84 posts
Gabriel
@ggaler7
Mostly Node.js/TS. Currently delving into everything Web.
Bucharest, RO Beigetreten Temmuz 2020
590 Folgt54 Follower
Gabriel retweetet
Gabriel retweetet
YouTube
Suomi
Gabriel retweetet
since all the frontier labs now offer "long context", 2026 is going be the year when everyone finds out how "long context" is actually implemented under the hood :)
Melkey@MelkeyDev
This is wild. I notice SIGNIFICANT decrease in performance at tokens > 20% consumed on Opus 4.6. It degrades INSANELY, like the 1M context doesn't matter. The model just starts being delusional and unusable. 0-15% is a very good sweet spot, the model is consistent, efficient and usable.
English
@mitsuhiko @1st1 Cool! Is this up on youtube? I hate watching videos here lol
English
Something different for the weekend. I talked with @1st1 about lat. He built a knowledge graph that connects specs with code and integrated it into pi and other harnesses. It's an interesting experiment.
English
PSA it takes max 30 minutes with claude code to get multi-head attention
Elliot Arledge@elliotarledge
less than 0.01% of the population understands how a transformer works
English
Gabriel retweetet
we as software engineers are becoming beholden to a handful of well funded corportations. while they are our "friends" now, that may change due to incentives. i'm very uncomfortable with that.
i believe we need to band together as a community and create a public, free to use repository of real-world (coding) agent sessions/traces. I want small labs, startups, and tinkerers to have access to the same data the big folks currently gobble up from all of us. So we, as a community, can do what e.g. Cursor does below, and take back a little bit of control again.
Who's with me?
cursor.com/blog/real-time…
English
a friend just sent me this and told me to stop crying
Gabriel Moncha@gabimoncha
i just opened claude and this happened @bcherny @jarredsumner pls, take this a bit more seriously
English
Gabriel retweetet
Gabriel retweetet
He bought the Mac Mini? Good. Now replace his feed with videos of people using Claude Code on the Vision Pro
English
@GeoffreyHuntley’s LP is a nice example x.com/geoffreyhuntle…
Maybe a business will be just an openclaw that manages the code and the data/analytics with a malleable interface for occasional human observation
The default view should be the one to help you make decisions
geoff@GeoffreyHuntley
English
I dream of an IDE where you just prompt what you want to see
me: hey, let’s work on this
ide: loads everything from logs, code, diagrams, user data concerning that problem space
An IDE embedded in the product but still flexible enough to generate the right UI
Andrej Karpathy@karpathy
Expectation: the age of the IDE is over Reality: we’re going to need a bigger IDE (imo). It just looks very different because humans now move upwards and program at a higher level - the basic unit of interest is not one file but one agent. It’s still programming.
English
Lots of non tech friends want openclaws. So far i've set them up on VMs, but this is getting heavy. Are there any good multi-tenant openclaw setups or alt-claws yet that are good enough?
English
Gabriel retweetet
~9 months ago I predicted the emergence of “open source, closed spec”
Many ai founders have said “I can screen share the code, I will never disclose our evals”
Now @tldraw
Malte Ubl@cramforce
Wow, @tldraw is moving their tests to a closed source repo to prevent a Slop Fork github.com/tldraw/tldraw/…
English
