Sabitlenmiş Tweet
Jaison
1.4K posts
Jaison
@jaisonsaji
ML Engineer • Computer Vision • മലയാളി • Retweets are bookmarks
observable universe Katılım Ağustos 2011
2K Takip Edilen627 Takipçiler
Jaison retweetledi
Talking about tracing things back
Flowers ☾@flowersslop
Every LLM from any lab today traces back to this guy, who was the only person at OpenAI pushing for pretraining transformer language models. He built GPT-1. After that did others see the potential. He invented it, and almost none of the so called AI experts even know his name.
English
Jaison retweetledi
You have built an entire identity around being someone who could succeed if they tried. That identity is a warm blanket. Real action is a cold shower. When you finally attempt something, you risk losing the story that has comforted you for years. What if you try and discover you are average? What if you try and find no special talent? This is not fear of failure. This is fear of ordinariness. So you stay in the planning phase forever. You buy the notebook. You watch the tutorial. You tell friends your idea. Anxiety spikes whenever someone asks for a deadline. You are not protecting your future. You are protecting a fantasy that has already expired.
WELCOME TO BLACK TWlTTER @blacktwiterthrd
Hit me with the harshest reality truth.
English
Jaison retweetledi
A lot of folks talk about "escaping the permanent underclass". If AGI pans out, the future class divide won't be based on wealth, but on cognitive agency. There will be a "focus class" (those who control their attention and actually do things) and a "slop class" (those whose reward loops are fully RL-managed by AI)
English
Jaison retweetledi
Billionaires are wasting money on anecdotally testing longevity hacks.
Here are the best longevity hacks that you learned in high school but forgot as an adult and then some.
Avoid tobacco.
Avoid alcohol.
3 cups unsweetened black coffee a day.
Adhere to vaccinations for age.
Take your booster shots too.
Predominantly plant based diet.
Minimum 8000 steps a day.
Min 154 min of aerobics/week.
Find and do work that makes you happy.
Keep and care for a pet.
Play video games.
Avoid bro science.
Avoid alternative medicine.
Follow real experts. Not role playing ones.
Listen to your doctor. Not the internet.
English
Stop everything you are doing and watch this. Fantastic video on how modern science works and how it evolved/evolves.
Krish Ashok@krishashok
Is Ayurveda a Science? youtu.be/xZ9ANYvIZG4?si…
English
Jaison retweetledi
Jaison retweetledi
we just released RF-DETR segmentation
SOTA real-time segmentation + Apache 2.0 license
six model sizes with performance spanning from 40.3 mAP at 3.4 ms/image (Nano) to 49.9 mAP at 21.8 ms/image (2XLarge)
link: github.com/roboflow/rf-de…
English
Jaison retweetledi
JEPA are finally easy to train end-to-end without any tricks!
Excited to introduce LeWorldModel: a stable, end-to-end JEPA that learns world models directly from pixels, no heuristics.
15M params, 1 GPU, and full planning <1 second.
📑: le-wm.github.io
English
@nkalra0123 yes, unfortunately that's a fear and an expectation.
everything that can be in the training distribution will eventually be in the training distribution
English
We found a task where LLMs struggle massively!
Give them a coding problem in Python and they'd work great. Give the same problem in brainfuck and zero-shot their performance is ~0%
+[--------->+<]>+.++[--->++<]>+.
Lossfunk@lossfunk
🚨 Shocking: Frontier LLMs score 85-95% on standard coding benchmarks. We gave them equivalent problems in languages they couldn't have memorized. They collapsed to 0-11%. Presenting EsoLang-Bench. Accepted to the Logical Reasoning and ICBINB workshops at ICLR 2026 🧵
English
Highest roi to date is to collect more data/clean up existing data.
Rohan Pandey@khoomeik
labs will publish details on arch, optim, objectives, scaling, kernels, literally everything except data and academia will be astounded for the hundredth time, wondering to itself where the secret sauce is
English
Jaison retweetledi
labs will publish details on arch, optim, objectives, scaling, kernels, literally everything except data
and academia will be astounded for the hundredth time, wondering to itself where the secret sauce is
English
This is super useful! I actually got it to generate a near perfect svg replica of a logo I created using nano banana pro (Gemini 3 pro and gpt 5.2 thinking failed to regenerate matching svg) Took a couple of tries but worked nevertheless.
QuiverAI@QuiverAI
Today we're opening our public beta access to Arrow 1.0 A first of it's kind SVG AI model. Turn your ideas into graphics.
English
1. Sarvam deserves all the praise.
2. It's big for India's AI ecosystem and I can't wait to check it out once they opensource the models, but can we really call it India's deepseek moment?
3. Galgotia fiasco was a disgrace to the event and overshadowed the good work sarvam and others did. No point in getting worked up when people point that out.
4. Opposition in India is always gonna be critical of the ruling gov. You think ruling party would've praised sarvam if they were opp?
neural nets.@cneuralnetwork
I'm not even a political person but WTF YOUR COUNTRY LITERALLY MADE A HUGE JUMP IN AI TODAY WE HAD A DEEPSEEK MOMENT But yes pls highlight the one moment out of 100000 moments where we failed I don't think this party is worthy to even be called an opposition, because the opposition should think for bettering the country never voting for them.
English
Been doing this for quite some time. Build and train models using pytorch, export to onnx and run on browser using onnx web runtime.
Check out jaison.in/move
AVB@neural_avb
I will train cool models that do cool shit, ship into my website(s), make them run on your browser at 50ms latency, and blow your fucking mind.
English