Frontend Enjoyer | Code | Football | DM for work YouTube - https://teapawt.net/t.co/RRY09Y1hub
Local AI
Joined July 2025
- Tweets2K
- Following219
- Followers178
- Likes4.5K
Aman Upadhyay retweeted
Andrej Karpathy:
“A 1B model can be much more effective than most current LLMs.
90% of current models waste a ton of capacity and money on remembering stuff that doesn’t matter.”
here is his 5 core thesis about Small LLM’s:
1. the intelligence was never the knowledge. Strip the memorized internet, keep the reasoning. What’s left is the cognitive core, and it’s small.
2. the core might fit in ~1B params. Maybe less. “Plenty of room at the bottom.” We’ve been paying for a library when we needed a brain.
3. size stopped being the bottleneck. What makes an agent work is the tools it can call and how tight the feedback loop is, not parameter count.
4. intelligence lives in the system, not the weights. A raw LLM is a CPU with no OS. Small core + good harness > giant model alone.
5. you can own it. A model on your hardware is one no one can rate-limit, price-hike, or deprecate on you.
in 12 months the gap between teams shipping on-device and teams paying per-token frontier bills won’t be capability.
TwIL an 8.8B model at 90%+ of frontier isn’t a benchmark flex.
It’s the thesis Karpathy’s been repeating all year, made real.
study more on this below ↓
Modest intelligence is arriving faster than most people realize.
Today, we're open-sourcing TwIL-8B.
Just 8.8 billion parameters, yet approaching frontier-model scores on several difficult benchmarks:
• 93% of frontier performance on coding
• 92% on instruction following
• 90% on contest mathematics
All from a model small enough to run on your own hardware.
But the benchmarks aren't the most important part of this announcement.
For years, the AI industry has focused on building the most capable model, regardless of the cost required to run it.
I believe that equation is changing.
The next era of AI will be defined by the cost of completing work. Not simply the intelligence of the model, but how efficiently that intelligence can accomplish the task.
If a model with 8 billion parameters can reliably complete the same work as one with hundreds of billions, why pay for the larger model?
As smaller models become more capable, intelligence becomes something we can distribute, specialize, and run across the hardware we already own.
That's the future we're building at @thewebAI
TwIL-8B is available today. Open weights.
And this is just the first of several announcements we're making today.
HF Link: huggingface.co/webAI-Officia…
Aman Upadhyay retweeted
Today @WSJ covered our work with @MercedesBenz. our flagship LFMs are now rolling out in vehicles in North America, running directly on the hardware inside the car to power in-car assistance! 🌊
this is one of the first deployments of local generative AI in the real world at scale! ✨
seeing our technology become part of someone’s everyday life is one of the most rewarding feelings of building Liquid.
years of work become a small, useful part of someone’s day. they don’t need to know which model is running. they just need it to understand them and work seamlessly.
this is really hard and it took us years of fundamental work from first principles to made happen with a brilliant team of engineers from Mercedes and Liquid.
I believe on-device AI will fundamentally change our relationship with the technology around us. embedded AI is about change of ownership.
With Mercedes-Benz, we are now in production and are building toward deployment at scale. The on-device AI revolution is taking shape in products people actually use, and I’m proud that @liquidai is helping lead it.
To our team and our partners at @MercedesBenzUSA and HQ: thank you for the care, persistence, and trust it takes to make this real.
It feels good to see the work out in the world.
wsj.com/cio-journal/mercedes…
Aman Upadhyay retweeted
#Dubai will soon allow travelers to apply for visas directly through the official Visit Dubai Portal.
Single and Multiple Entry Tourist Visas can be applied. More categories will be added in the future.
visitdubai.com/en/plan-your-…
You can register your interest to get notified once they start the same.
#DakuVisaUpdates
Aman Upadhyay retweeted
Comunicado Oficial: Rafael Nadal.
realmadrid.com/es-ES/noticia…
Aman Upadhyay retweeted
Rotos de dolor, con un nudo en la garganta y sin poder creer que esto sea verdad, desde el CD Tenerife comunicamos el trágico fallecimiento, este martes, día 3 de noviembre de 2026, a las 8:00 horas, de Jesús de Miguel Alameda.
No hay palabras de consuelo; no hay absolutamente nada que mitigue este sentimiento repleto de tristeza. Estamos junto a su desolada familia, junto a sus seres queridos, junto a todo el tinerfeñismo que hoy llora por este cruel desenlace.
La familia de Jesús de Miguel agradece todas y cada una de las muestras de apoyo recibidas, e igualmente solicita, de corazón, máximo respeto a su privacidad, en estos terribles momentos.
🖤 DEP
Aman Upadhyay retweeted
15 saal pehle patri pe hagte they aur aaj bullet train ki manufacturing ho rahi. 2029 haare toh laanat hai humpe
Aman Upadhyay retweeted
aight, fuck it, my turn for this kind of post
i'm looking for work.
i'm 21 and i've spent a rather unreasonable amount of time building software over the past year.
here's some of what i've been up to:
kiban: built an embedded LSM storage engine from scratch in rust. WAL, SSTables, compaction, snapshots, concurrency, crash recovery and fault injection.
the benchmarks are probably my favourite part. on warm 1M-record workloads, i got 0.96µs median point reads and over 1M point-read calls/s, with 2.2-2.7x faster point reads than tuned RocksDB. group commit pushed fully durable writes to 5,379/s at 16 writers, a 7.5x improvement. also managed to shrink the point index by 61%.
styx: a ~41k LOC bittorrent client in rust with 600+ tests. implemented peer-wire networking, verified downloading/seeding, trackers, DHT, magnet metadata exchange, daemon IPC, CLI/TUI and a desktop shell.
kurogane: implemented Raft consensus from scratch in rust. elections, log replication, snapshots, crash recovery, joint consensus, real gRPC networking and a deterministic fault-injecting simulator with linearizability verification.
konto: built a double-entry accounting engine with PostgreSQL and typescript. atomic transactions, holds, transfers, idempotency and enforced financial invariants. includes stress testing with 1,000 concurrent transfer/hold operations.
iris: built a GPU-accelerated image viewer for linux using rust, Vulkan, GTK4 and DMA-BUF. RAW decoding, ICC colour management, GPU rendering, Wayland presentation and software fallbacks.
webnotes: built a local-first notes app with a native Tauri desktop client, SQLite storage, bidirectional linking, a structured editor and web synchronization.
scribe: built an open-source publishing SDK for React, Next.js and Vite. handles Markdown/MDX compilation, typography, components, styling and local authoring. currently in public beta.
and then there's my open source work.
i've had PRs merged into HelixDB, Turso, Apache DataFusion, Zed, NativeLink, Zilliz Knowhere and Nix, covering everything from database internals and query execution to infrastructure, editor behaviour and correctness fixes.
i've also got contributions underway in RocksDB, Neon, HydraDB, FalkorDB and Starship, among others.
most of my work revolves around rust, typescript, databases, storage, networking and systems engineering. i've also been getting deeper into ML infrastructure and inference.
i'm looking for a full-time remote engineering role. internships and contracts are on the table too.
my github has all the code, benchmarks and documentation:
github.com/aetosdios27
if you're hiring or know someone who is, hit me up.
reposts appreciated :)
Aman Upadhyay retweeted
I did say we'd be releasing our models periodically...introducing Gemini 4 Carbon, our next frontier model.
Following Argon's release just two weeks ago, Carbon shows the pace at which we’re pushing the frontier, it ranks #1 on the recently updated v5.0 of @ArtificialAnlys Intelligence Index at 59. It also leads across key cybersecurity evaluations scoring 88.2% on SEC-bench Pro for finding vulnerabilities, 54.8% on ExploitGym for exploitation, and 71.0% on CWE-bench for patching.
We’re seeing significant improvements in long-horizon tasks even compared to Argon, and we’ve strengthened frontier safeguards.
As we did with Argon, we’re going to roll it out gradually + responsibly. Some trusted cyber defenders will get access starting this week. We'll be expanding it to more partners through our Fairwind Program, ahead of wider availability soon, so hang tight!
Aman Upadhyay retweeted
Introducing 𝙻𝙵𝙼𝟹: our new generation of Liquid Foundation Models. With a new tri-hybrid architecture and new multimodal encoders, the LFM3 family is the ideal model suite for specialized intelligence.
> 6 model instances, across text, audio, vision, and omni modalities, generative and decision models.
> 𝙻𝙵𝙼𝟹-𝟷.𝟽𝙱: text, 256k context window.
> 𝙻𝙵𝙼𝟹-𝚅𝙻-𝟸𝙱: vision-language model with video-input support for spatial understanding
> 𝙻𝙵𝙼𝟹-𝙰𝚞𝚍𝚒𝚘-𝟸𝙱: tool calls, voice chat, reasoning in 18 languages.
> 𝙻𝙵𝙼𝟹-𝙾𝚖𝚗𝚒-𝟸𝙱: an omni model combines vision, audio, and language.
> 𝚍𝟷-𝟸𝙱 and 𝚍𝟷-𝙾𝚖𝚗𝚒-𝟸𝙱: decision models with typed outputs and calibrated probabilities in a single pass. Omni adds audio input.
> fine-tune on 1 GPU.
> a new tri-hybrid architecture
> architecture optimized for 14 processor classes across GPUs, CPUs, and NPUs.
> 𝙻𝙵𝙼𝟹-𝚅𝙻-𝟸𝙱 is a beast. It performs competitively with Gemma 4 12B (6x larger), and Gemini 3.5 Flash Lite, and GPT-5.6 Luna, with 1.9B parameters.
> 𝙻𝙵𝙼𝟹-𝟷.𝟽𝙱 is #1 on the Pareto frontier of device intelligence 𝚙𝚒𝚙𝚎𝚝𝚝𝚎.𝚕𝚒𝚚𝚞𝚒𝚍.𝚊𝚒
Every modality for specialized intelligence. Pick the one that fits your use case, customize it, and run it on-device.
1/ 🧵
Aman Upadhyay retweeted
Mobile PLI stats as follows:
Government incentive paid: Rs 19,000 Cr.
Direct tax collected: Rs 25,000 Cr.
GST collected: Rs 3,00,000 Cr.
Electronics manufacturing employs 25 Lakh people; mobile manufacturing alone employs 12 Lakh.
Cope harder!!
One in four iPhones is made in India, but imported parts and costly subsidies complicate the success story. @andymukherjee70 takes apart an iPhone to explain why copying China's manufacturing model may not pay off 🎥
Aman Upadhyay retweeted
Guys,one of my childhood friends has started Makhana business,these are the six flavours,I have personally tasted all flavours and they are really amazing,best part is,it's gluten free and roasted.
He wants to expand his business all over India and also in international markets.
If anyone of you knows any snacks distributor/exporter in your city or if you are into same field then kindly reach out to me,will share further details with you. Thank you🙏🏼✨❤️
Aman Upadhyay retweeted
whenever @NICKIMINAJ trends our web traffic goes up
Aman Upadhyay retweeted
#WATCH | Berlin, Germany | Union Minister Piyush Goyal says, "Who's responsible for this mess in the first place? Where did the ozone layer get damaged? Who sent all that carbon into the atmosphere?... I represent 17% of the world's population, but my contribution to global carbon emissions is less than 3%. You (Europe) represent probably 3% or 4% of the world's population, yet your contribution to those carbon emissions—or this mess—is humongous...The problem was caused by you...Even today, as we sit here, my per capita contribution to carbon emissions is probably less than 10% or 15% of yours. Even today, Europe's per capita contribution to carbon emissions—and its per capita environmental impact—is probably seven or eight times that of India...I'm representing 17% of the world's population, and I am more conscious of waste and the circular economy. I also have to meet the needs of 1.4 billion people. I have to meet their energy needs, their food needs, and ensure that my nation remains secure. Despite all of that, we have one of the world's fastest-growing clean energy programmes..."
(Source: Piyush Goyal Social Media Page)
