Founder @ Ordalie | https://teapawt.org/t.co/1tcHhyBaxW
Paris, France
Joined February 2019
- Tweets2K
- Following1.5K
- Followers592
- Likes19.9K
It's fantastic, can't recommend it enough
no matter what agent you're using, your voice is the best way to talk to them
and hex is the best way to use your voice:
- fast
- runs locally
- works on mac, windows, linux
stable release now available: anoma.ly/hex
everyone talks about vibe coding, nobody talks about vibe testing
Agents can fuzz your app way harder than any of us would ever bother to by hand. Weird inputs, edge cases, broken payloads..
That's what actually makes vibe coded software stable. Writing code got cheap, but so did breaking it
Anthropic rewrote its Messages API in Rust. Zero Rust devs on the team.
Claude wrote the code. Humans wrote the plan and checked old vs new matched on real traffic. CPU dropped 40-50%.
Pretty sure "never do the big rewrite" is dead advice.
@katelyn_lesse on @latentspacepod
It feels like nobody pays for a benchmark score anymore
Grok4.7 is far from the top of Artificial Analysis index and lots of people are crazy about it, massive adoption with no free tier.
That's the plateau showing- for everyday work even mid tier models are enough and the value has moved to the harness, the tools/context/integrations
Ofc raw intelligence still matters where expertise runs deep like research or hard science. Everywhere else you're paying for the harness, not the model
ok this is the best agent setup i've had so far.
a Grokbot chief of staff runs our @buzzdotxyz. it opens channels itself, briefs the other agents, follows up with them, and keeps the whole company in one context.
i'm no longer the one who has to remember what everyone is doing, feels so futuristic and works amazingly well.
COS doesn't need to be as smart as opus, but it prompts my intents better than i do
Since the last update, hex is absolutely flying
As a long wisprflow customer i can guarantee you there's no need to pay for stt anymore. This thing is crazy, really crazy.
just set it to whisper-large-v3-turbo and insert your domain terms in the "hint" field. 1gb of ram max, and faster than wisprflow since the latest update
github.com/anomalyco/hex
Models like astra and opus 5.5 are the equivalent of money falling from the sky and the goal is to cast the biggest possible nest to capture the upside
For now the biggest net we found is @buzzdotxyz most of you are missing out honestly
I also think people like having choices made for them. Choice is cognitive load, and figuring out what to ask a black box to get what you want is tedious for a lot of people.
A big % of tech people don't get it
Normies won't ask "build me software"
They will ask "I have this photo and I want to manually edit it" and it will spin up a Photoshop clone
They will ask "I need to know if sales are going up or down this quarter from all different ecommerce platforms we use" and it will spin up an analytics dashboard connecting all APIs with scheduled crons that keeps it up to date forever
They already do!
They won't know they're building software because their AI app/agent will do it for them
And then they'll stop using most regular software!
Chrome extensions are so underused rn. With coding agents any websites just becomes a backend you can exploit to your will with your own interface. More on this sooon
true! same as the 3D demos: they know exactly what will make a great showcase on X and bruteforce those specific capabilities into the weights.
Pretty much every motion video demo I see here follows the same patterns.
Still super impressive though!
Introducing Notion in the terminal
I spend most of my day in @herdrdev and I want to stay there. My agents already work great with Notion but I had no good way to follow along next to them. So I built ntty
It sits on top of the excellent Notion CLI from @NotionDevs, and it covers most of what I do day to day. Still a few rough edges and kanban views aren't there yet. That's next.
PRs are welcome :)
Baudouin retweeted
After working on it for a year, I can finally see the end state of the harness
On the backend, it's an execution environment that gives the model maximum degrees of freedom and on the frontend it's a free-form canvas that the model can use to communicate with the user
The only valid argument for post-training inside a vertical is cutting costs in tier 2 tasks.
Everything else is marketing or capital incineration
Pretty interesting post.
Max was quite critical of Harvey doing post training and has gone on podcasts saying this. This was also used as a point of differentiation for Legora where they don’t do post training and just focus on building the wrapper layer. I agreed with the points he made.
From the post it seems like they are going to do post training now. We are in 2023 all over again.