building https://teapawt.net/t.co/XUILyw7Wei / acquired @ 20
New York
Joined May 2017
- Tweets242
- Following98
- Followers227
- Likes607
Elijah Popowitz retweeted
Thank you everyone for the support!
Since this post reached 50k views, next 50 people whose first deposit is >$10 get 50% bonus credits
(Up to $50 bonus credits)
Sign up with this link: trymidas.io/r/LAUNCH50
or
In Dashboard, go to Billing
Code: LAUNCH50
$10 deposit = $15 in credits
...
$100 deposit = $150 in credits
We achieved SOTA on Terminal Bench 2.1 using a DeepSeek v4.1 Flash model.
92.2% at $0.08 cost per task.
Previously Opus 5.5 was the leader at 87.6% at $0.50 cost per task (vals.ai/benchmarks/terminal-…).
Now the obvious question is how did we do it? How can you possibly improve a Flash model to beat frontier models like Opus, Fable, and Astra?
The truth is, we over-obsess about models when they are only half of the equation. The other half is the code that wraps around the model. In fact all of the most revolutionary products (Instinct, Muse, Claude Code, OpenClaw, etc.) are all just harnesses optimizations.
Rather than try to build a new model or fine-tune, we built a harness that would make:
Flash Models —> Frontier Results
Frontier Models —> Beyond Frontier Results
Did I mention it’s also ~6x faster than Claude Code?
And we’re now live. Go Check it out!
Trymidas.io
Elijah Popowitz retweeted
We also achieved SOTA on @CommandCodeAI's Frontier Token Efficiency eval.
This benchmark measures read and shell tool calling token-efficiency and we crushed.
Read more:
trymidas.io/benchmarks/token…
(Yes we will run TB4 in the next couple weeks)
Since this post went viral, next 50 people whose first deposit is >$10 get 50% bonus credits up to $50.
$10 deposit = $15 in credits
...
$100 deposit = $150 in credits
In Dashboard, go to Billing
Code: LAUNCH50
Elijah Popowitz retweeted
We achieved SOTA on Terminal Bench 2.1 using a DeepSeek v4.1 Flash model.
92.2% at $0.08 cost per task.
Previously Opus 5.5 was the leader at 87.6% at $0.50 cost per task (vals.ai/benchmarks/terminal-…).
Now the obvious question is how did we do it? How can you possibly improve a Flash model to beat frontier models like Opus, Fable, and Astra?
The truth is, we over-obsess about models when they are only half of the equation. The other half is the code that wraps around the model. In fact all of the most revolutionary products (Instinct, Muse, Claude Code, OpenClaw, etc.) are all just harnesses optimizations.
Rather than try to build a new model or fine-tune, we built a harness that would make:
Flash Models —> Frontier Results
Frontier Models —> Beyond Frontier Results
Did I mention it’s also ~6x faster than Claude Code?
And we’re now live. Go Check it out!
Trymidas.io
Elijah Popowitz retweeted
I'm hiring founding research scientists and interns for @MorphLabs_ to build non-invasive brain interfaces.
Looking for exceptional researchers from any background to tackle SSL on thousands of hours of EEG, transfer learning across people, sessions and headsets, and causal decoders that adapt online while the user adapts back
Elijah Popowitz retweeted
I think we are quickly evolving past needing surfaces to review intent/alignment and into needing better decision making "command centers."
As decision making across many contexts, which agents effortlessly own, is becoming the bottleneck.
This underscores the need for knowledgeable decision makers – and less so about a bottleneck of us not understanding; better models are able to derive aligned intent and produce tasteful+correct solutions without too much steering. This enables the experienced and things are just starting to get fun as we babysit less and do less of "wait bro wtf did you just do?"
UI/UX:
Similar to intent/alignment surfaces, and perhaps ironically bc better decision making surfaces need to provide adequate context per decision (that is easy for a human to consume/understand).
e.g. similar things, but different:
A tool feels behind when it focuses on a side bar of message thread activity and inline html rendering vs a tool that manages multiple contexts, surfaces relevant html artifacts per thread, and abstracts the exhaustive message threads away from you.
Elijah Popowitz retweeted
My analyst when the investment committee asks him a question he should absolutely know the answer to
Absolute masterclass.
Take notes.
Elijah Popowitz retweeted
@karpathy highlighted the benefit of a personal tutor during his appearance on the @dwarkeshpodcast podcast perfectly:
The ability to figure out exactly where their pupil's skill level is and challenge them just enough to push their understanding further.
I love teaching and eventually want to open a brick and mortar school myself post @morphlabs_ai days. Until then, the goal is to figure out how to deliver that personal tutor relationship at scale.
I'm starting a new ongoing project to try and replicate that experience. It's called microtutor, with the intention of helping anyone gain a deep understanding of topics they desire. Constantly tracking your mastery and evaluating your understanding of topics.
I'm releasing the first version here. It's a cli application, still rough around the edges, but I'm looking forward to having people try out the experience.
github.com/s-mehra/microtuto…
If you try it, tell me what's broken and what could be better. That's the only way this kind of education will get better.
