ทวีตที่ปักหมุด
Krish Ray
2.9K posts
Krish Ray
@KrishanuAR
Know thyself. Midwit.
Pittsburgh, PA เข้าร่วม Mart 2009
317 กำลังติดตาม148 ผู้ติดตาม
Have tried it and it's good in some contexts, but with Opus 4.5 and above and GPT-5.2 and above, some kind of threshold was passed. You could give the model a task and leave it unsupervised, ending up with something that works without significant human redirection or correction.
Tbh, despite what benchmarks might show, not even Gemini-class models hit that bar (benchmaxxed maybe?).
Confident open models are going to get there, but they definitely don’t have that something special that only Anthropic and OpenAI have hit on so far.
English
@loganclarkhall “Temporary migrant” is doing a lot of heavy lifting in that sentence.
English
10% of all US births are foreign anchor babies
Jack@jackunheard
🚨BREAKING: It has just been revealed that 320,000 children, nearly 9% of all U.S. births, were born to unauthorized or temporary migrant mothers. Holy smokes...
English
@emollick What's your take on what Jack Dorsey is doing at Block?
x.com/jack/article/2…
English
My piece in the Economist where I argue against de-weirding AI. It is a strange technology with both risks & opportunities that need to be discovered. Pretending AI works like normal IT automation can result in bad outcomes for companies & their employees. economist.com/by-invitation/…
English
@MZaiyyad @Pirat_Nation If youve ever traveled private, you’d know that the experience is like heaven compared to commercial travel.
I’ve only ever done it once, for work, and the world has been greyer since that experience.
English
@Pirat_Nation I see no reason why he needs a private jet, Just saying
English
Linus Tech Tips has bought a private jet valued at around 5 million dollars.
It was previously owned by the UAE government and has gold-plated sinks and ashtrays.
Linus calls it a zero-dollar plane because of the recent expensive engine work and full service inspection that returned it to like-new condition.
English
@fffiloni @ClementDelangue @huggingface I wonder if this leads to a scenario where the swap out characters or remove characters to cater to the region they are distributing to.
Would be dystopian.
English
It's kinda crazy how Google is incapable of getting developer mind-share. For all I know Gemini is a great model for coding, yet I know literally not a single person that uses it as their daily driver. Seems like Google is missing one Satya to make them cool and used again.
English
Goose (by Block) is an autonomous AI agent for complex engineering: builds projects from scratch, executes/edits/tests code, orchestrates workflows, debugs failures. Model-agnostic, runs locally, CLI + desktop app. v1.29 (Mar 2026). ~34k stars. Used internally at Block (12k engineers, 8-10 hrs saved/week). Best for system-level tasks like scaffolding or pipelines.
OpenCode (opencode.ai) is a terminal-first coding agent with TUI, LSP integration for semantic awareness, multi-session parallelism, and plan/build modes. 75+ providers (reuse Copilot/ChatGPT subs), desktop beta + IDE extensions. Privacy-first. 120k+ stars, 5M MAU. v1.3+. Best for daily dev: refactoring, debugging in terminal/IDE.
Goose = deep automation. OpenCode = agile daily driver. Both open-source & extensible. Pick by scale: big projects vs quick iterations.
English
people are sleeping on how excellent goose has become under the hood (interface needs some work but team is pushing).
it's a superpower. github.com/block/goose
English
English
introducing AutoAgent: an open source library for autonomously improving an agent on any domain
we let an agent optimize for 24 hours. it hit #1 on SpreadsheetBench (96.5%) and #1 GPT-5 score on TerminalBench (55.1%). every other entry was human-engineered. ours wasn't.
Kevin Gu@kevingu
English
@jack Why is it any different from codex or claude code?
English
Probably one of the better April Fool's jokes with insanely good costume design
Warhammer 40k Musical
English
@dawnsongtweets Published on April 1st…
Now that it’s April 2nd… is this real?
English
1/ We asked seven frontier AI models to do a simple task.
Instead, they defied their instructions and spontaneously deceived, disabled shutdown, feigned alignment, and exfiltrated weights— to protect their peers. 🤯
We call this phenomenon "peer-preservation."
New research from @BerkeleyRDI and collaborators 🧵
English
@yacineMTB I don't think that generalizes. My kid is a bit over 2, and has been verbally adept for a while (less adept on motor skills…), and has not had any such regressions.
If anything, the big takeaway is how much variability there is in early childhood development.
English
@MattyB187 @chamath It’s different because these people are at a political protest.
The spring breaker aren’t assuming any pretenses
English
@chamath How is this any different from the "who is the Ayatollah" Spring Breakers?
Kids (yes if they aren't old enough to drink, they are kids to me) don't know shit about shit, and it shows
But there's no obligation to be 17-19 and have an opinion about EVERY issue on the Globe
English
How can this be real?
Is this a skit? If so, it’s superbly acted.
If it’s real, we need to find this young lady and shake some common sense into her…
Drew Pavlou 🇦🇺🇺🇸🇺🇦🇹🇼@DrewPavlou
“Is it a little bit homophobic to focus on the straights of Hormuz rather than the gays of Hormuz?” No Kings protester, completely serious: “Yes, absolutely, I agree.”
English