Just launched on Product Hunt and got 52 upvotes. Could you guys help some more? :) producthunt.com/products/may…
2
3
75
After months of training models, we are launching Maybloc today. It is a SOTA for speech-to-text for overlap intensive audios. And it has the lowest errors among Pyannote Precision 3, Deepgram Nova-3, and Assembly AI 3.5Pro.
Shoot you messiest audios at it on maybloc.com!
2
54
jev's most important application would be in the next gen harnesses
system 1, jev-like model
system 2, llms
2
66
ah yes, anything to steal more research and call it your own
xitter.gq/Linahuaa/status/209729…
Demand for Astra is really unprecedented. We're pulling all the levers possible to sustain the demand, but I've not seen anything like it until now and we went through very steep growth before. Priority will always be to keep excellent service for existing users, but we might have to pause new Pro subscriptions for a bit if this continues.
2
179
> The proof was produced by a group of agents
unethically stealing someone else's solution who used your product a few times is the new normal
Their Head of Ethics, Head of Safety Systems, Head of Data Centers and a dozen others all left in the LAST 3 MONTHS
They are always reading and digging all chats to find research to steal that they can't do themselves. Even Satya Nadella said this openly
No wonder they gave $2M to all yc companies to steal their IP
We’re sharing a solution to the Navier-Stokes Millennium Prize Problem, one of the deepest problems at the frontier of mathematics.
The proof was produced by a group of agents, using an OpenAI next-generation model significantly more capable than GPT-6 Astra.
The problem concerns whether the description of smooth three-dimensional fluid motion modeled by the Navier-Stokes equations can break down. It has remained unresolved for roughly 90 years.
2
79
hes right in some sense but mostly projecting his skill issue
I've worked with and seen some debian devs know even tiniest details of large projects they build/maintain and some even used by 1M+ users/agents a day
this industry has the most disloyal customer base I've ever seen.
All strategies of burning cash for MAU will fail inevitably if there's no better anchor to offer
in the past week, i've used for more than 2 hours each:
- Codex
- Claude Code
- OpenCode
- OpenCode Desktop
- Pi
- Cursor
- Devin
- Grok Bot
- T3 Code
it's a very turbulent time to main AI coding tools
2
93
Replying to @reasonmachines
@aradotso Can you guys fix this?
It’s broken beyond use. You guys have been adding features and confetti, but what a coding agent needs first and foremost is reliability and basic usability. Neither of which it currently has.
1
60
llm's first few tokens decide the whole response
I've seen several QA dataset answers leading with the conclusion first and then the explanation which is simply stupid and puts a lot of effort on those first tokens
llms are autoregressive, they need to lead to the conclusions through yapping
if the first few token get the conclusion wrong, the rest of them will just spend tokens justifying it (aka hallucination, but in a much subtle yet extremely common form)
1
1
47
I wasted 2 months researching and testing papers to make RAG more efficient and fast.
At the end nothing came out of it. The industry is already neck deep and happy with mid but working solutions.
I guess 2 hours of market research could have save me those 2 months.
But I would also haven't found a better problem which I'm pivoting to now.
I'm no one to give suggestions but I've learnt the hard way that the only way to find better problems is to just start. Start working on mid problems, test market, talk to people and pivot quickly.. several times. Cliché
32
I'm giving away tryhenka.com for free.
I don't have time to manage/market it and my interests keep shifting: models rn
lmk if you/anyone wants it
2
3
59














