Akash Goswami

Build something awesome!

[Akash Goswami @aagosh]

@ishaanpreet @SkyrootA Thesis here is that Indian tourists will go to space with Haldiram in their backpacks.

[Etched @Etched]

Introducing Low-Voltage Inference (LVI) for high throughput workloads.

Today, AI chips can't scale FLOPs without thermal throttling.

As FLOPs utilization increases, AI chips draw more power and downregulate clock speed. This often results in sustained inference throughput under half of peak FLOPs.

Chips in other industries solve the power problem by running at lower voltages. Bitcoin miners run at under 3x the voltage of AI chips!

We’ve designed a new architecture to run our chip’s math blocks at under half the voltage of most AI chips. This enables multiple times the FLOPs density of AI chips today.

We can run trillion parameter sparse MoEs at 80%+ peak FLOPs without thermal throttling.

Running LVI requires co-designing the entire cluster from the transistor to the token: new splittable math arrays, circuit techniques, novel tiling and scheduling algorithms, power delivery networks, VRM architectures, advanced packaging, cold plate designs, and more.

[Akash Goswami @aagosh]

Getting feedback on your live website is often a struggle. I launched a small javascript library which turns your live website into a Figma style review board.

https://t.co/70RJ29UAs0

Appreciate feedback from Web Authors!

[Akash Goswami @aagosh]

@AnthropicAI First it was AI hardware, now it is Software. We are seeing weaponization of AI tools by the governments.

Anyone looking to deploy models and must now think about sovereignty and self governance of infrastructure.

[Akash Goswami @aagosh]

@elonmusk, now that Tesla Model X and S are discontinued, why not open source their entire design and firmware?

I'm pretty sure existing owners would love to see continued support for parts/accessories and firmware update beyond it’s mandatory lifetime.

[Akash Goswami @aagosh]

@igorhansachat This tweet summarizes the sorry state of EU’s affair. Not my job to check the link worked - says the intern who posted this.

[Unsloth AI @UnslothAI]

@googlegemma Thank you Google Deepmind for constantly releasing open models! 🌟

We made Dynamic GGUFs so you can run Gemma 4 12B more efficiently: https://t.co/8cL321pVDh

[Akash Goswami @aagosh]

@hcompany_ai Holo3 has been my browser vision workhorse. Super excited to try H3.1

[H @hcompany_ai]

Computer-use agents are moving from the cloud to your local machine. Fast.

When we launched Holo3 two months ago, the production feedback was clear: digital agents need to be blazing fast, cost-effective, and versatile.

Today, we're dropping Holo 3.1, engineered to run anywhere, instantly.

Massive token throughput. Low latency. Ready for your local workflow!

[MiniMax (official) @MiniMax_AI]

Introducing MiniMax M3: The First Open-Weights Model to Combine Three Frontier Capabilities

- Coding & Agentic Frontier: 59.0% SWE-Bench Pro, 66.0% Terminal Bench 2.1, 34.8% SWE-fficiency, 28.8% KernelBench Hard, 74.2% MCP Atlas
- MiniMax Sparse Attention scales context to 1M
- Natively Multimodal from Step Zero

API: https://t.co/fHRdSV7BwZ
Token Plan: https://t.co/BDCycxepZw
🚀New! MiniMax Code: https://t.co/GvB4YiB6Ul

Weights & Tech Report in ~10 Days

[Akash Goswami @aagosh]

@thsottiaux cli must have a /usage command

[Akash Goswami @aagosh]

@antirez ds4 is the best local model for my mac studio m3 ultra overall, however I keep switching to qwen 3.6 for speed. A possible improvement in decode speed will
make it super awesome

[Akash Goswami @aagosh]

@thsottiaux I am using codex and claud side by side on the same project and same repo. They review other's work and build on top, while writing handoff guides.

hard to say but both are equally impressive in their own ways. The peer review also makes their effectiveness multiplied

[Akash Goswami @aagosh]

@Prince_Canuma this is the craziest model i have tested so far. 185 t/s on m3 ultra 256 gb

Thanks for the work you are doing!

[Akash Goswami @aagosh]

@AlxTurovski @justinhammon_ Nice — there's definitely a market. I've been building LogSonic (open source, MIT) for log analysis — might be worth looking at how I handle the Go backend + Bleve search + MCP server for AI agent integration. Always good to know the landscape!

[Akash Goswami @aagosh]

@Prince_Canuma cant wait

[Akash Goswami @aagosh]

@elonmusk you are leaving existing X owners high and dry, especially in Europe where there is no FSD and Cybertruck

[Akash Goswami @aagosh]

Zurich is buzzing with Physical AI talent. Here is nice summary from City of Zurich of all the companies in the area https://t.co/cXvt4rFIQb

[Akash Goswami @aagosh]

@antirez’s ds4.c made me realize that weight is not a product, inference is a product.

Expect more hyper optimized inference engines for specific LLMs in the future.

[Akash Goswami @aagosh]

@googleaidevs Curious why validation of drafter tokens is faster than token generation itself.

If that’s true, can’t we use CPU for speculative decoding, assisting GPU even more?

Or even use large trie datasets to do speculative decoding instead?