Austin Hirsch

Sharing my AI implementation journey, tailored to business owners

Owner of in-house metal and plastic part manufacturing + Ecomm Brands | 100% USA Made

[Austin Hirsch @austinhirsch]

https://t.co/QagAmltQT7

[Andreas Storm @avstorm]

Uber has introduced a new icon system

[Unsloth AI @UnslothAI]

We’re releasing new Qwen3.8-27B GGUFs with 10% higher accuracy.

Unsloth Dynamic V3 outperforms others by >10% on Div-300, KLD & more benchmarks.

We also release 1-bit quants that retain 77% accuracy. Run on 8GB RAM.

Blog: https://t.co/tHsBexyh2K
GGUF: https://t.co/xIdNwm7CLQ

[NVIDIA AI @NVIDIAAI]

We benchmarked 300+ NVIDIA verified skills to see how much they actually help agents on real tasks.

Same task, same model, same setup. The only difference was whether the agent had the skill.

Across the benchmarks, skills improved correctness by 41 points, effectiveness by 39, and efficiency by 35.

SkillEvaluator is open source if you want to test your own skills before you ship them.

[Asad Memon @_asadmemon]

It is done! Mighty’s Batch-2 is ready to ship today! ✅

Mighty is the smallest VIO/SLAM module. Here are the specs and features:

- Global shutter camera + IMU
- 0.8W peak power
- Outputs pose at 20Hz; IMU runs at 800Hz
- Scalable on-device loop closure
- Occupancy mapping through the SDK
- Multi-camera synchronization via FSYNC
- Connects to drones, Arduinos, ESP32s, phones, RPis, and more
- Powered by 5V USB or 3–15V input
- Lua programmable for custom protocols and behaviors

Orders are open now worldwide.

[Jamie Barton 🥕 @notrab]

Setup a new Mac. Gone full Apple Container instead of Docker.

This app makes it so much nicer https://t.co/KVbQnsZPQT 🍏

[Jakub Antalik @Jakubantalik]

Made a skill for UI transitions:
It reviews a project, identifies transitions, and suggests improvements or replacements based on the source Transitions-dev library, making them more polished and snappy.

npx skills add Jakubantalik/transitions.dev

[Liquid AI @liquidai]

Today, we release DSpark draft models for LFM2.5-1.2B-Instruct, LFM2.5-2.6B, and LFM2.5-8B-A1B. These add a speculative decoding path that trades a minimal memory increase for a large decoding speedup without changing output quality.

A lightweight draft model proposes a block of candidate tokens and the target model verifies them in a single forward pass. Across MATH500, GSM8K, HumanEval, MBPP, and MT-Bench at batch size 1:

> Up to 3.18x throughput on an H100: LFM2.5-8B-A1B on MATH500, 428 → 1362 tok/s
> Up to 2.87x on an M4 Max MacBook Pro: LFM2.5-1.2B-Instruct on HumanEval, 136 → 389 tok/s
> LFM2.5-2.6B means: 2.67x on the H100 (323 → 864 tok/s), 2.27x on device (61 → 139 tok/s)
> Under greedy decoding, the emitted sequence is identical to baseline by construction, so benchmark accuracy is unchanged.

Each draft model is around 300M parameters, with embedding and LM head tied to its target model. The gain shows most in agentic workloads, where the model reasons before every tool call and the user waits through it all: on BFCL multi-tool scenarios, DSpark cuts LFM2.5-2.6B latency by nearly 50% on average.

🧵

[Pi @pidotdev]

A harness turns a model into an agent. At it’s core it provides 4 things:

- a system prompt
- tools
- an agentic loop
- a translation layer across models

New blog post from Earendil co-founder @colindaymond on what a harness is, and how you can own yours. Full post below

[Chester @chesterzelaya]

OpenHD, an open-source project that turns off-the-shelf Wi-Fi chips into hardware for long-range, low-latency video transmission

It holds spot #7 for long-distance flights, with a previous world record of 55 km!

https://t.co/hQqgViYYOJ

[Datapoint AI @datapointai]

today, we’re releasing the largest open-source human image preferences dataset, along with a $1 million data grant

- 2M+ annotations by real people
- 30 SOTA image models ranked
- 10 categories (marketing, product design, anime etc)

dataset + benchmark + grant details below:

[BlackwellBoy @Blackwellboy]

Qwen 3.8 27B is not one number. Same weights, wildly different results depending on reasoning, runtime, context, quant and serving stack. Several days of testing across 5 DGX Sparks + RTX 5090 https://t.co/l20qWfc263

[Alex Cheema @alexocheema]

Excited to collab with @ashxhart on this!

Why does prefill/decode disaggregation work on DGX Spark + Mac ?

DGX Spark: ~350 TFLOPS@fp4, 128GB@273GB/s
M5 Max MacBook Pro: ~70 TFLOPS@fp4, 128GB@614GB/s

Prefill is compute-bound. Prefill on Spark runs ~5x faster than M5 Max.
Decode is memory-bound. Decode on M5 Max runs ~2x faster than Spark.

Simple idea: run prefill on Spark, decode on M5 Max. That's faster than running on one device alone, since each phase runs on the best device for that phase.

This requires sending the KV cache from the Spark to the Mac. With a 10 Gigabit connection, we can overlap fully computation and communication of the KV cache. This doesn't require low latency, and works just fine over TCP/IP.

RDMA on the other hand enables low latency networking (~single digit microseconds), which is the unlock for scaling up decode with ~equal memory bw devices e.g. DGX Spark + M4 Pro / M5 Pro.

Low-latency is also important for DGX Spark + RTX or Mac + RTX. Small pool of very high speed memory + large pool of slower memory is the ideal local hardware architecture. Idea is to put the attention part of the model on the faster small memory pool, and the MoE part on the slower large memory pool.

[Akshay 🚀 @akshay_pachaar]

Sam Altman made the case for open-source harnesses in July.

a month later, someone shipped it, and it's more efficient than most managed harnesses.

here is the problem it was aimed at:

a large share of your agent's token bill is the model rereading things it already read.

that isn't the model's doing. the runtime around it decides what goes into every prompt and how often the model gets called.

for example, an agent queries a CRM at step four and gets back 400 rows. those rows get piled up in the conversation history.

by step nineteen, the model has to read those rows fifteen times unnecessarily, and every token read is billed at input rates.

it happened because your harness assembled that prompt on every turn and kept the rows in it.

that gives you two levers: how much context the harness carries forward, and how often it calls the model.

there are four practical ways to keep the prompt from growing unnecessarily:

→ load tool schemas on demand. a server with 100 tools doesn't need to put all 100 into every prompt when the agent only calls two.

→ offload large results to disk. turn a large response into a short preview and a file path instead of replaying the entire result on every turn.

→ delegate to subagents. let a subagent spend thirty tool calls in its own context and return one summary to the root agent.

→ run toolchains in code. one script calls three tools, joins the results, and returns a table instead of three turns each dragging a full response.

but reducing context is only half the job. you also need to control how often the model gets called.

a good harness should avoid unnecessary planning, verification, and reflection when the work can be completed in fewer steps.

@TrueFoundry's open-source agent harness, TrueForge, is built around both of those controls.

it sits between the model and the tools, deciding what goes into every prompt and when another model call is actually needed. it also breaks token usage down across the harness, skills, instructions, tools, and messages.

DevRev's Enterprise-Bench is where this gets tested, on multi-step tasks of the kind where an agent pulls records from one system and reconciles them against another.

TrueFoundry ran TrueForge there against Claude Managed Agents, both on the same model, and both finished the same number of tasks.

the tie is the part that matters, because it means the gap underneath is not a quality tradeoff.

TrueForge reached that score on close to a third of the tokens, with roughly 40% fewer trips back to the model. for the same result, that comes out around 2.7x cheaper than Claude Managed Agents.

swapping in an open model made it sharper still. TrueForge with GLM-5.2 scored a little higher than either setup above, and the entire benchmark run cost about $3 at list prices.

being open source matters beyond the license here. the model underneath can be swapped without rewriting the agent, and the whole thing can run inside your own environment when the data cannot leave it.

all of this comes down to the runtime around the model, the context it carries, the tools it exposes, and how many times it goes back to the model.

that is what a production harness actually owns.

the full task list, the per-run numbers, and the MIT-licensed code are on GitHub: https://t.co/Ueo1InZH7M

(don't forget to star 🌟)

you can read more about the same in the article quoted below.

thanks to the TrueForge team for working with me on this one.

[Austin Hirsch @austinhirsch]

Cost of power and raw materials needs to be inversely correlated with artificial intelligence

If we can do this, abundance will be for everyone.

If we don’t make this a priority today, we will create Elysium on accident

[Ornith @ornith_]

Aloha! 🌺Introducing Ornith-1.5, a family of open-source LLMs spanning 9B Dense, 35B MoE, and 397B MoE, trained with self-improving strategies.

It achieves state-of-the-art performance among open-source models of comparable size and delivers performance comparable to Claude Opus 4.8 across reasoning, agentic, and coding tasks:
✅Terminal-Bench 2.1 (86.1)
✅SWE-Bench (86 on verified, 65.1 on pro, 79.6 on Multilingual)
✅DeepSWE (56)
✅HLE (44.6)
✅ClawEval (81.4)
✅Tool Decathlon (71.2)

Ornith-1.5 takes a major step toward training foundation models through end-to-end self-improvement, extending the self-scaffolding strategies introduced in Ornith-1.0 into a more complete self-improvement loop: the model proposes new tasks, generates task-specific scaffolds, and produces solution rollouts for reinforcement learning, continuously creating new learning experiences from which it can improve.

All models, along with their quantized versions (FP8, GGUF, MLX, and NVFP4), have been released under the MIT License, enabling unrestricted commercial and research use.
📘Tech Blog: https://t.co/OZ63scRWLB
🤗Huggingface: https://t.co/mGJLwhrQOM

[Austin Hirsch @austinhirsch]

@scottstts It’ll just be phone hardware. Models will be tiny and they’ll work on this generation iPhones.

[OpenMed @OpenMed_AI]

We gave LFM2.5-VL-3B from @LiquidAI a generated skin-like image and let it choose tools.

It mapped 6 regions, drew the contours, measured L04 at 8.8 × 8.1 mm, then picked what to review first.

Local on a Mac Studio. Visual review, not diagnosis. 🧵

[0xSero @0xSero]

We got computer use at home.

GLM drives this so well, I’ve been using it day and night and having it manage my computer and invoices.

Free and open source

[Austin Hirsch @austinhirsch]

I don’t understand how this makes sense...

I thought this was a ai joke

[Karen Bass @KarenBassLA]

When you donate to our campaign, the city can match your donation 6 to 1! If you donate $5, the city turns it into $35. Our campaign is powered by YOU!
Click the link in our bio to make your matching donation today!

https://t.co/shTvSCpGGg

[Konrad Reczko @reczko_konrad]

Depth-aware light injection in TypeGPU

I got a 448x448 monocular depth model down to ~8 ms on my M4 Pro across ~250 dispatches, which is fast enough to use in realtime :D
Since the inference is written directly in TypeGPU, I can just feed the depth buffer straight into the lighting pass. It never has to leave the GPU or go through any extra synchronization/interop step

Inference, lighting and draw all go through the same command encoder.