Simon Edwardsson

CTO & Co-founder @v7labs
Building reliable GenAI agents at scale (V7 Go)
Small models • finetuning • agent UX • TTS • CLI • CV
Experiments →

[Simon Edwardsson @SimEdw]

Trained a tiny 9M Conformer-CTC model to finally fix my terrible Mandarin tones. Strict per-syllable grading (no helpful 'corrections'), runs 100% in-browser (~11MB).

Huge thanks HN, woke up to 250 points on my Show HN!

[V7 @V7Labs]

V7 Go now connects to all your favorite apps via MCP Integrations

→ 2–3 click setup
→ Agents pick the right actions on its own
→ Results that look native to each app, not just a wall of text
→ Works at massive scale across agent tables

Learn more: https://t.co/XLShBQ4GRT

[Nathan Benaich @nathanbenaich]

News! @airstreet has raised $232,323,232 for Fund III to back AI-first companies from the earliest stages in the US and Europe.

Now the largest solo GP venture firm in Europe.

Our third epoch begins today. Join us!

[Cheng Lou @_chenglou]

I’m very happy to present my toy research project: Sotaku!

It's a neural net that automatically discovered the rules of sudoku and learned to solve them, achieving a new state-of-the-art score of 98.9% on one of the hardest sudoku datasets, while being agnostic to the game, and beating all other sudoku-optimized neural net architectures*

Read more for fun motivations, plus some extremely unconventional discoveries, e.g. reverse curriculum consistently beating curriculum (!), emergent reasoning-like capabilities, and the future of traditional programming

[Simon Edwardsson @SimEdw]

Just wrote the ultimate productivity killer:
A Magic Eye screensaver where crossing your eyes reveals the real time.

✅ Unreadable by passersby (and Zoom calls)
✅ Instant 90s flashback
✅ Guaranteed "is your GPU dying?" comments

Can you read it?

[Simon Edwardsson @SimEdw]

@KeridwenCodet @patloeber It’s unlikely that 3.1 was trained from scratch, probably used continue to train on 3.0

[Simon Edwardsson @SimEdw]

@KeridwenCodet @patloeber The fewer models the need to host the easier it’s to maintain, scale etc. Have a lot of different models is very suboptimal

[Simon Edwardsson @SimEdw]

NVIDIA’s venture strategy is basically: Here’s some cash… now go buy our GPUs with it.

[Simon Edwardsson @SimEdw]

Most teams I talk to have the same problem: one person figured out how to use AI really well, and nobody else knows how.

The workflow lives in their head. It doesn't spread.

We just shipped something at @V7Labs to fix this, have a look.

[Simon Edwardsson @SimEdw]

@gpt_alex That's a nice project, did something similar for mandarin but trained a small CTC model, maybe you could try something similar? https://t.co/2cdjAFah1r

[Simon Edwardsson @SimEdw]

Poll’s closed and honestly the results line up with what I’m hearing everywhere.

I still try to do a little "analog coding" each week, keeps the brain from completely atrophying...

[Simon Edwardsson @SimEdw]

Asked founders & CTOs at dinner:
"When did you last write a full function without AI?"

Most: months ago 👀
You?

Poll ↓

[Simon Edwardsson @SimEdw]

This is wild: https://t.co/4J8Znw96ag ~15K tokens/sec 🤯

Custom silicon running Llama 3.1 faster than any GPUs.

When this hits newer models, whole new categories of apps unlock.

[Simon Edwardsson @SimEdw]

Asked founders & CTOs at dinner:
"When did you last write a full function without AI?"

Most: months ago 👀
You?

Poll ↓

[Simon Edwardsson @SimEdw]

@conductor_build Emojis caused it to fail at first, but later any character in certain chats caused it to fail, and I emailed you a Loom video yesterday with more details

[Simon Edwardsson @SimEdw]

If @conductor_build shows a grey screen when you type certain characters into the prompt, it’s likely a QuotaExceededError from localStorage.

Fix: press ⌥⌘I → clear localStorage cache → refresh.

Took me way too long to figure out 😅

[Simon Edwardsson @SimEdw]

@OfficialLoganK Does this mean that 3.0 Pro will stay preview forever in favour of 3.1?

[Simon Edwardsson @SimEdw]

Audio input’s a little rough right now. Fixing that next.

[Simon Edwardsson @SimEdw]

First hackathon of the year ✅

Built a voice-driven game remix app for kids, honestly one of the most fun projects I’ve made in a while.

Thanks for hosting @CircleCI and @AITinkerers!

[Simon Edwardsson @SimEdw]

Under the hood: voice → LLM → code edits → instant game updates

Built with Elixir LiveView + Gemini Flash.
It can ask follow-up questions before modifying the TypeScript.
If it passes typecheck, it ships live.

[Simon Edwardsson @SimEdw]

@claudeai Is this the first time a Claude model has launched on Anthropic’s API without also being available on Vertex AI or Bedrock?

[Simon Edwardsson @SimEdw]

@sama Overall, I think this is good news for OpenClaw: more resources and no longer a need to find a way to monetise.