manavv
@manavvnotop
Followers
127
Following
9K
Media
70
Statuses
794
super30 @100xDevs • applied LLM mechanic • developer
Pune
Joined August 2024
things i am betting on in 2026: - memory/context engineering for AI applications and AI agents - small language models, i can already see some releases - voice AI is next big thing intuitively, i can already see some interesting things happening in the voice space what do you
2
0
10
this is some incredible stuff
⚡ Faster than Fast. Designed for Agentic AI. Introducing Xiaomi MiMo-V2-Flash — our new open-source MoE model: 309B total params, 15B active. Blazing speed meets frontier performance. 🔥 Highlights: 🏗️ Hybrid Attention: 5:1 interleaved 128-window SWA + Global | 256K context 📈
0
0
2
i said it. voice models are the next big thing
This is the DeepSeek moment for Voice AI. Today we’re releasing Chatterbox Turbo — our state-of-the-art MIT licensed voice model that beats ElevenLabs Turbo and Cartesia Sonic 3! We’re finally removing the trade-offs that have held voice AI back. Fast models sound robotic.
0
0
3
this is some reaaally reaaally good stuff
Introducing Bolmo, a new family of byte-level language models built by "byteifying" our open Olmo 3—and to our knowledge, the first fully open byte-level LM to match or surpass SOTA subword models across a wide range of tasks. 🧵
0
0
4
huge props to @MiniMax__AI for this. they charge you based on the number of prompts you give it, not on the basis of number of api calls a single prompt does to a model.
0
0
1
too good “why i still write raw code instead of prompting the compiler”
Someone asked Gemini to imagine HackerNews frontpage 10 years in the future from now:
0
0
1
don't build agents, build skills instead. a great talk by @barry_zyj & @MaheshMurag, about how we can improve the performance of agents by giving them skills.
2
0
7
sometimes you gotta hop off multiplayer and start grinding story mode. now, i am almost at the final checkpoint.
0
0
2
some insane stuff happening right now. great motivation for me to learn fast and ship faster.
You can now train LLMs 3× faster with no accuracy loss, via our new RoPE and MLP kernels. Our Triton kernels plus smart auto packing delivers ~3× faster training & 30% less VRAM vs optimized FA3 setups. Train Qwen3-4B 3x faster on just 3.9GB VRAM. Blog: https://t.co/j3HpsmzrCw
0
0
5
survived air india, my nostrils are now finally breathing in 4K quality at home
1
0
7
this is huge from @MistralAI i told you, small language models excelling in a niche is the new big thing
Our next-generation coding model family Devstral 2 is available in two sizes: Devstral 2 (123B) under a modified MIT license, and Devstral Small (24B) under Apache 2.0. Both SOTA, open-source, free to use, and available now via our API.
0
0
6
well, if anyone is looking for a AI engineer, they can hire me 😉
talked about some open source models with @manavvnotop . reduced the cost to 10x from gpt, anthropic to open source models. now he is asking for party :)
1
0
9
all i need is a boiler room set playing in the background to get the fuck locked in.
0
0
1
a good resource to learn about agents btw
📢 A must-complete course for end of year 5-Day self-paced Agents Intensive Course by Google https://t.co/QxhrYaOwGw 🔖
0
0
2
“desiring of positive experience is itself a negative experience, and the acceptance of a negative experience is itself a positive experience.” - mark manson
0
0
5