Himanshu retweeted
We're seeing more and more teams use data generated by agents to improve their agent platforms.
My notes on how coding agents, evaluations, and shared infrastructure make self-improving agent platforms possible.
agno.com/blog/towards-s…
My notes on how coding agents, evaluations, and shared infrastructure make self-improving agent platforms possible.
agno.com/blog/towards-s…
10
10
52
2,726
Himanshu retweeted
🌟 Excited to share that our paper, RoboReason, has been accepted to EMNLP 2026!
🤖 We study how well Vision-Language Models (VLMs) perform on spatiotemporal ordering tasks that use real-world robotics images.
Our key findings: 🔹 Large VLMs often underperform previously valuated on spatiotemporal ordering tasks using real-world robotics images. 🔹 Overlaying specialized computer vision model outputs and increasing reasoning effort generally help improve large VLM performance. 🔹 Finetuning smaller models using GRPO and SFT on examples generated by specialized computer vision models often helps them achieve comparable results to much larger VLMs.
📍 Presenting with @sagnikanupam at @emnlpmeeting in Budapest 🇭🇺 tomorrow!
🗓️ Poster Session C | 9:00–10:30 AM
#EMNLP2026 #VLMs #LLMs #Robotics #AI
🤖 We study how well Vision-Language Models (VLMs) perform on spatiotemporal ordering tasks that use real-world robotics images.
Our key findings: 🔹 Large VLMs often underperform previously valuated on spatiotemporal ordering tasks using real-world robotics images. 🔹 Overlaying specialized computer vision model outputs and increasing reasoning effort generally help improve large VLM performance. 🔹 Finetuning smaller models using GRPO and SFT on examples generated by specialized computer vision models often helps them achieve comparable results to much larger VLMs.
📍 Presenting with @sagnikanupam at @emnlpmeeting in Budapest 🇭🇺 tomorrow!
🗓️ Poster Session C | 9:00–10:30 AM
#EMNLP2026 #VLMs #LLMs #Robotics #AI
19
21
127
8,294
Himanshu retweeted
Introducing Orcus.
No LLM API key required to make AI model requests.
Orcus uses your web credentials to access AI models available through supported provider websites and deliver responses to your applications through a familiar, API-style interface.
From model invocation and outputs, tool calling, and batch processing, Orcus brings essential AI integration capabilities into one developer-friendly SDK.
Explore Orcus: orcus.site/docs/gemini
No LLM API key required to make AI model requests.
Orcus uses your web credentials to access AI models available through supported provider websites and deliver responses to your applications through a familiar, API-style interface.
From model invocation and outputs, tool calling, and batch processing, Orcus brings essential AI integration capabilities into one developer-friendly SDK.
Explore Orcus: orcus.site/docs/gemini
15
24
78
5,338
Himanshu retweeted
the best score on swe marathon isn't opus 5.5 or gpt-6 astra.
it's an open-weights model you can download, at a fifth of astra's price.
kimi k3.1 pro: 58.3
opus 5.5: 46.3
astra: 42.5
$2 / $10 per million tokens vs astra's $10 / $50. 1m context. plus a highspeed version, 3x n't wait.
that's $40 back on every million tokens out.
it's an open-weights model you can download, at a fifth of astra's price.
kimi k3.1 pro: 58.3
opus 5.5: 46.3
astra: 42.5
$2 / $10 per million tokens vs astra's $10 / $50. 1m context. plus a highspeed version, 3x n't wait.
that's $40 back on every million tokens out.
Introducing Kimi K3.1 Pro: our most capable 3T-class open-weight model.
🔹 1M-token context, 2.8T total / 104B active params (MoE), Multimodal.
🔹 Full-project coding: games in Unreal, Unity & Godot.
🔹 Excels at parametric 3D modeling in Blender & CAD.
🔹 More reliable at finance with hallucination down by 65%.
🔹 Matching the strongest proprietary models across coding, agentic and knowledge benchmarks.
Kimi K3.1 Pro and K3.1 Pro Highspeed (2–3× faster) are now live on Kimi.ai, Kimi Work, Kimi Code, and the Kimi API.
Open weights by Nov 4, 2026.
🔗 API: platform.kimi.ai
🔗 Tech blog: kimi.ai/blog/kimi-k3-1…
7
5
20
1,190
Himanshu retweeted
this article ended up as a pr to sglang-omni, and sometimes making inference fast isnt always about writing the fastest GEMM, there's still lots of other stages you can work on! x.com/i/article/2111…
7
5
84
2,386
Himanshu retweeted
It’s done. Pac-Man can now play against Laya.
Laya reads the game state and decides Pac-Man’s next move in real time.
4 ghosts. Classic maze. One AI-controlled Pac-Man.
Everything runs locally.
Built with an open-source decision model.
Here’s Laya in action.
Laya reads the game state and decides Pac-Man’s next move in real time.
4 ghosts. Classic maze. One AI-controlled Pac-Man.
Everything runs locally.
Built with an open-source decision model.
Here’s Laya in action.
2
2
2
81
Himanshu retweeted
How it works: scripted workloads stress a Linux box (CPU/memory/disk abuse + innocent bystander app), a rules "teacher" records what it would do, and the model learns to pick from 12 resource actions with a safety layer: it can never kill or restart anything on its own.
0
2
2
59
Planning to start the old habit of building different projects for myself, but been disconnected for too long has made me less creative, drop some ideas which you think could be solved with the help of
"Agentic inference" because i am planning to make a project out of this intersection.
and if you are interested we can work together as well.
COMMENT below 👇
"Agentic inference" because i am planning to make a project out of this intersection.
and if you are interested we can work together as well.
COMMENT below 👇
8
0
18
756
A chill space(tech is partially allowed ) x.com/i/spaces/1PKqr…
0
0
4
286
Himanshu retweeted
Build the setup that makes people ask, “How the hell did you do all that alone?” x.com/i/article/2090…
15
7
38
10,902
Himanshu retweeted
Think Quant Memo is just quant finance? Think again.
Algorithmic trading. Machine learning. Deep learning. Neural nets and LLMs. Stochastic calculus. Derivatives. Market making. And the research shaping all of it. One place. All free.
We're here to disrupt, and we're coming ls.
Something new ships every week. Take what they've been charging you for 👉 quantmemo.com
Algorithmic trading. Machine learning. Deep learning. Neural nets and LLMs. Stochastic calculus. Derivatives. Market making. And the research shaping all of it. One place. All free.
We're here to disrupt, and we're coming ls.
Something new ships every week. Take what they've been charging you for 👉 quantmemo.com
110
136
585
388,101
As a brown kid, I don’t know if it is too late or early to understand about a few things in life.(last one is something which we all have done)
- make sure no other parameters are above your health & family’s health.
- its good to earn & have passion but spend time with your family
- you can always be surpassed
- you need to control your passion & curiosity to make it an eternal flame rather than burning it continuously.
- make sure no other parameters are above your health & family’s health.
- its good to earn & have passion but spend time with your family
- you can always be surpassed
- you need to control your passion & curiosity to make it an eternal flame rather than burning it continuously.
4
1
26
1,112
My mom said something to me a few days back & it changed my pov.
“you stay in home but you don’t”
It was like an eye opener for me.
Maybe it’s time we realize we don’t want our parents waiting for the same love we used to ask from them when we were kids and needed them.
“you stay in home but you don’t”
It was like an eye opener for me.
Maybe it’s time we realize we don’t want our parents waiting for the same love we used to ask from them when we were kids and needed them.
0
0
3
262
Himanshu retweeted
I built a second brain to cut my token bill.
It didn’t.
A 2026 study found LLM-written context files cost 20% more, with no measurable gain. Your agent can already read the repo.
It doesn’t know what you decided, what failed, or what must not change.
That’s what I carry harness has to learn it twice.
It didn’t.
A 2026 study found LLM-written context files cost 20% more, with no measurable gain. Your agent can already read the repo.
It doesn’t know what you decided, what failed, or what must not change.
That’s what I carry harness has to learn it twice.
switch from claude code to codex mid-task and lose nothing. move semantic state, never KV state x.com/i/article/2083…
21
12
49
7,432
Himanshu retweeted
Grok Bot shouldn't write the code.
It should run the team that does.
I gave mine Claude Code, Codex, OpenCode, and pi.
You define the outcome.
It decides who plans, who builds, who reviews, and when you need to step in.
One interface.
An army of coding agents.
Let it route rewrite what has already been decided.
It should run the team that does.
I gave mine Claude Code, Codex, OpenCode, and pi.
You define the outcome.
It decides who plans, who builds, who reviews, and when you need to step in.
One interface.
An army of coding agents.
Let it route rewrite what has already been decided.
switch from claude code to codex mid-task and lose nothing. move semantic state, never KV state x.com/i/article/2083…
31
13
145
21,861
Himanshu retweeted
I have five coding harnesses installed. Cursor, Claude Code, Codex, OpenCode, Prime Agent.
I still can't tell you which one deserves which task.
Claude Code will spin up a swarm of sub-agents to rename a variable. pi will bring four tools to do a job that needed the swarm.
So setup, and every prompt lands on whichever one is actually best at it
I still can't tell you which one deserves which task.
Claude Code will spin up a swarm of sub-agents to rename a variable. pi will bring four tools to do a job that needed the swarm.
So setup, and every prompt lands on whichever one is actually best at it
switch from claude code to codex mid-task and lose nothing. move semantic state, never KV state x.com/i/article/2083…
16
20
146
26,528