Adi (@devsvn2) header
Adi (@devsvn2) profile picture

co-founder https://teapawt.org/t.co/wYzIXKGWKe (1.4K+ waitlist) | Applied ai @ https://teapawt.org/t.co/VvLOjbTVdN | prev https://teapawt.org/t.co/blt4XKSX6k | I am good at basketball

London, England
Joined February 2022
Photo by Adi (@devsvn2)Photo by Adi (@devsvn2)Photo by Adi (@devsvn2)Photo by Adi (@devsvn2)Photo by Adi (@devsvn2)Photo by Adi (@devsvn2)Photo by Adi (@devsvn2)Photo by Adi (@devsvn2)Photo by Adi (@devsvn2)Photo by Adi (@devsvn2)
The lore behind Microsoft’s recent outage Get ready to see this in your next system design interview lol
89
394
50
9,667
1,116,459
People still pay 10k for cohorts promising “live sessions, hands on projects and industry mentorship” Just ask ChatGPT or Claude to teach you with a structured plan. Find project ideas from engineers who regularly share their work here and build them yourself There’s really no reason to spend 10k on cohorts these days lol
6
2
1
17
795
Jensen Huang will make his comeback as NVIDIA CEO in 2027
8
1
23
858
Jensen Huang resigning as NVIDIA CEO is the biggest plot twist of 2026
10
1
29
1,072
Qwen 4 Max preview, What's with the wheels ? My dear God, physics oriented models need to be in the mainstream. Day by day @ylecun 's world model makes sense to me. The more a model understands casuality, order, & how physical world works. I'm not a aware of the architecture they have used for this, but most mainstream models are sloppy anyway. The harness is the one making them do some capable things (that's again is the brain of an engineer making a worst "half baked" product to perform to its fullest potential)
2
4
6
334
Stop being so cynical about your agents. Trust them, build, ship and hit that $10K MRR faster
6
1
20
836
If you want to stay on top of the latest tech trends while working, joining a funded startup is probably your best bet
9
2
40
1,402
Adi retweeted
Fable 5.2 one shotted this btw
Introducing Claude Fable 5.2 and Claude Mythos 5.2, designed to tackle the most daunting technical challenges. They’re capable collaborators on ambitious projects in many fields, and show improvements in areas like coding, reasoning, and vision.
48
21
5
639
292,712
Stat always exaggerate things, but Anthropic is the only company where the models have consistent good quality. Unlike some other AI companies.
Introducing Claude Fable 5.2 and Claude Mythos 5.2, designed to tackle the most daunting technical challenges. They’re capable collaborators on ambitious projects in many fields, and show improvements in areas like coding, reasoning, and vision.
1
2
3
52
why is standard llm inference slow? plain autoregressive decoding makes k tokens cost k serial runs of the model - every token waits for the last one. speculative decoding breaks the queue: a small draft model guesses several tokens ahead, and the big model scores them all in one parallel pass. accepted guesses count; the first rejected token gets resampled from a corrective distribution, so the output distribution stays exactly the big model's. no retraining, no architecture change. on t5-xxl they reported 2-3x faster inference with the distribution unchanged. most tokens are easy - don't spend the big model's serial time on them until they need checking.
5
3
10
288
Happy diwali!
4
6
855
If he had asked even one tech person on X before posting this, he could’ve saved himself What are you even trying to sell man? This isn’t YouTube. People here can see through this crap
It costs ₹800-₹1500 to buy one domain name for a year. We are giving that for FREE. And you get 100+ AI models for chat, image generation, video generation, slides, deep research, website building in one platform. All for just ₹999 a month :)
5
1
14
2,436
1
1
14
526
Adi retweeted
Indian company interview assignment: Our company cost of LLM is drastically increased Build 105B llm and benchmark against opus5.5 1 week timeline, internet allowed
3
1
4
267
My llm is production ready.
2
3
9
225
yesterday turned out way better than we expected. crossed 100k views and 12k site visits with people from 129 countries checking out Engage and US driving the most traffic @devsvn2 and I couldn't have asked for a better start. huge credit to the team behind this
Anthropic’s second door for people who don’t get in tryengage.si
7
4
1
17
29,027
We at tryengage.si are trying to save students' time and get hired fast.
We just killed Notion, and open-sourced it. Introducing Terminus Notes: everything you like in Notion, removed the stupid AI credits and confusing sidebars. 100% Free, 100% Open source, 100% Customizable. You can export your data or point it to your Claude at any time.
4
6
610
Making hiring feel a little more human. That's the idea👀
Anthropic’s second door for people who don’t get in tryengage.si
1
2
2
100
Adi retweeted
Sarvam is as distracted as me. Just that it is funded and I am not!
Today we’re releasing Sarvam Prism, a System One model built on Sarvam 105B for fast, accurate decisions across enterprise workflows. It scores 71.8 with a median response time of 44 milliseconds, the highest average score among decision models that respond in under 100 milliseconds.
6
2
1
57
3,340
Anthropic’s second door for people who don’t get in tryengage.si
Applications are now open for the Anthropic Residency, a paid, fixed-term research program for academics, technical founders, and staff-level engineers who want to make AI safe and beneficial for everyone. No ML background is required. Apply by November 22: alignment.anthropic.com/2026…
18
38
5
986
112,176