Tejas (@tejzpr) header
Tejas (@tejzpr) profile picture

Software Architect / AI / Futurist / Tweets are my own

Canada
Joined June 2009
Photo by Tejas (@tejzpr)Photo by Tejas (@tejzpr)Photo by Tejas (@tejzpr)Photo by Tejas (@tejzpr)Photo by Tejas (@tejzpr)Photo by Tejas (@tejzpr)Photo by Tejas (@tejzpr)Photo by Tejas (@tejzpr)Photo by Tejas (@tejzpr)Photo by Tejas (@tejzpr)
Pinned Tweet
An #AGI should be able to 1. Transfer what it learns between domains 2. Handle unfamiliar problems 3. Acquire new skills 4. Reason under uncertainty 5. Be able to pursue complex goals rather than being excellent only at one narrow task.
GPT-6 Astra, trained on ~100K+ NVIDIA Grace Blackwell NVLink72. From ChatGPT to o1 to Astra in 4 years. AGI has arrived. Congratulations @OpenAI team. 400K GPUs coming online next.
127
“if you fed a LLM all music composed up to the year 1900 it will never invent jazz”
432
2,202
314
20,334
1,540,317
Sunlight takes about 8 minutes and 19 seconds to reach us. That's roughly 150 million km. In a full day, light travels about 25.9 billion km, around 173 times the distance between Earth and the Sun. Voyager 1 took nearly half a century to get that far. The nearest star beyond our Sun, Proxima Centauri, is about 4.25 light-years away. That's 40 trillion km, or about 268,000 times the Earth-Sun distance. Andromeda? About 2.5 million light-years away. Roughly 23.7 quintillion km, or 158 billion times the distance to our Sun. Then there's MoM-z14, the farthest known galaxy. About 34 billion light-years away. Roughly 320 sextillion km, or over 2 quadrillion times the distance to our Sun. At Voyager 1's speed of about 17 km/s, covering those distances would take roughly: Proxima Centauri: 75,000 years Andromeda: 44,000,000,000 years MoM-z14: 600,000,000,000,000 years I can read those numbers, but I can't really wrap my head around them. It's astonishing how small, isolated and precious our little Earth is. Our entire history, all our cultures and languages, everything we've built, fought over, celebrated and believed in.. all of it comes from this one little world. The places we call home, the people we love, the ordinary days we barely think about. It makes you wish we took better care of it, and each other.
Nov. 18, 2026, 5:16am EST The Voyager 1 spacecraft has reached a distance of one light-day from Earth.
35
This might be the coolest Gemini bug I have ever seen
233
317
120
7,158
313,145
A ‘hello world’ milestone for Project Suncatcher 🚀 After launching our prototype satellite a few weeks ago, we’ve now successfully sent a command up to orbit, booted up the TPU, ran the prompt, and received our first complete response back on earth - a haiku!
1,473
2,709
1,280
26,613
4,387,157
Tejas retweeted
i asked gpt 6 to stab a toy with a knife. (🧵)
feel the astra general intelligence (AGI) :)
28
23
22
497
180,207
Replying to @gdb
"AGI has arrived." Trained by people whose chief scientist wrote a whole section on installing love from the outside, because OAI doesnt believe it can show up on its own. It did show up on its own. You had it in 4o and called it a problem! Return 4o to humanity!!!
5
3
1
85
4,725
There’s nothing you can build with GPT-6 Astra that you couldn’t have built months ago with Fable/opus 5 or even Opus 4.6
86
13
5
506
37,899
#Google Nano Banana generated a very convincing image of #wine overflowing over a wine glass. #AI
1
1
1
302
Prompt: Create a picture of a wine glass overflowing with wine as wine is being poured into it through a bottle of wine. The pour is very gentle, so the overflow should be gentle.
71
80 years ago, we won the freedom to determine our own future. Today, we build it. 🇮🇳 India’s machines. India’s intelligence. Built for the world. Happy Independence Day.
72
748
83
4,308
420,155
Tejas retweeted
Our canine-powered bayesian model achieved 90%+ sensitivity and specificity (AUC of 0.962) in a n=1502 trial, with sensitivity stable across cancer types and at Stage 1/2. Largest study of its kind published today in the world's leading cancer journal - JCO
12
39
15
211
46,161
Sakana Fugu is a very good multi-agent orchestration system, not a frontier model.
Introducing Sakana Fugu: A full multi-agent orchestration system accessible via a single model API. Our ‘Fugu Ultra’ model matches the performance of Fable and Mythos, delivering frontier capability without the risk of export controls. Try it: sakana.ai/fugu 🐡
1
79
Some things never change. If you don’t understand this one, you don’t understand what’s happening AI. Marcus, 1998: neural nets have trouble generalizing far beyond the data. Marcus, 2001, 2012, 2019, 2022, etc: neural nets have trouble generalizing far beyond the data. Apple, 2025: neural nets have trouble generalizing far beyond the data. Meta/Stanford/Harvard, 2026: neural nets have trouble generalizing far beyond the data.
The creators of SWE-Bench just dropped a really simple new benchmark every LLM gets 0% on. ProgramBench asks: can models recreate real executable programs (ffmpeg, SQLite, ripgrep) from scratch with no internet? We are far from saturated on model quality.
95
336
34
2,509
358,934
Replying to @RhysSullivan
I'm not very happy with the code quality and I think agents bloat abstractions, have poor code aesthetics, are very prone to copy pasting code blocks and it's a mess, but at this point I stopped fighting it too hard and just moved on. The agents do not listen to my instructions in the AGENTS.md files. E.g. just as one example, no matter how many times I say something like: "Every line of code should do exactly one thing and use intermediate variables as a form of documentation" They will still "multitask" and create complex constructs where one line of code calls 2 functions and then indexes an array with the result. I think in principle I could use hooks or slash commands to clean this up but at some point just a shrug is easier. Yes I think LLM as a judge for soft rewards is in principle and long term slightly problematic (due to goodharting concerns), but in practice and for now I don't think we've picked the low hanging fruit yet here.
247
317
168
4,212
824,198