Figuring out!
Joined October 2022
- Tweets804
- Following292
- Followers238
- Likes1.1K
Pinned Tweet
as part of @AWS strands agents hackathon - Agents for humans, I built something I wished existed - a companion in your browser which can help you understand things visually.
Alongside - a chrome extension with a whiteboard you can ask visuals about anything.
A quick demo:
while making models to not be so overconfident, they made models under confident.
whenever asked will this work - oh I am not sure. it may not. okay let me spend 1B more tokens.
1B tokens later - oh no, it has these flaws, I fixed them
okay will it work now?
umm not sure.
codex @OpenAI decided to stop in between in a 4 hour long research task saying flagged prompt
did main agent prompt one of 8 subagents in invalid way or how did u realize this 4 hours later?
now can't continue the session, total waste!!
Tried out Muse Spark Open at visual grounding tasks.
OMG!!! It's super impressive.
At times performs much better than Muse Spark 1.3 itself.
Finds out even tiniest defects in a defects in cookware benchmark I am running.
Great job @finkd, @alexandr_wang and team!!
Today we’re releasing Muse Spark Open, the most capable US open-weight model to date.
Built for coding and agentic work, it excels at long-horizon tasks, complex software engineering, and multi-step reasoning.
Key Model Specs:
→ 816B total / 42B active parameters
→ 1M token context window
→ Natively multimodal across text, image, video, audio
Harpinder Singh retweeted
Harpinder could have joined a big company. He chose a small team where things were “still messy enough.”
A BITS Pilani graduate, he wanted a real problem to solve and room to make a meaningful contribution. He found Oximy on YC and sent a cold email.
What convinced him to join, and what did he find when he got here?
Meet Harpinder Singh, Founding Engineer at Oximy.
Glad you’re building with us, Harpinder (@singhhcoder) 💛
Got access to typesafe Jev and built a scout for deals.
It takes deals happening across platforms like Amazon Myntra etc, and verifies against my criteria.
Uses noul for probability and choice for reasoning classification, and sends alerts for only matches with high confidence.
After co-inventing ChatGPT, I kept asking myself: why have superhuman chat models not led to AGI?
I’ve spent the last 2 years in stealth building a new way to train models (RLCD), and a new type of frontier AI model that we are releasing today: Jev
• 20-200x faster
• 40-400x cheaper (w/ output tokens free)
• Frontier composable intelligence optimized for decisions
AFAICT the shortest path to AI-based economic revolution
Built gates of galloo for @OpenAI webmcp challenge.
Due to currently ongoing openai outage, couldn't submit it last minute
@OpenAIDevs pls consider the submission 🙏
@dot_js @gao_jude @justin_rushing
my gdoc homescreen right now
thanks to all @ycombinator alums and others who were so kind enough to review our application
Best part about applying to @ycombinator ?
It forces you to introspect, and articulate in simplest words what you are doing, why you are doing it, and for whom?
This is the skill you anyways need to master as a founder to hire, and to sell your thing.
when you forget to do the cleanup 🤣
didn't expect from a 5 year old established product
(this is from their play store description page)
We built something at @qordinate_ai that we are absolutely not comfortable releasing.
It's called Qordinate Mythos.
Internally, it let assistants coordinate with other assistants and reduced wasted coordination time by 70%.
No joke.
Also possibly the last normal sentence in this post.
At first we thought we had built a productivity system.
What we may have actually built is a coordination organism.
So no, we are not doing a public launch.
We are letting in only 11 people under Project QWing.
If you get access, that's on you.
If you don't, Mythos has judged you.
Check if you made it: qordinate.ai/mythos
