Search: “engineering”

It's a drop-in. One program_id field and it plugs into your existing engine configs such as KV offloading + speculative decoding. Format-agnostic by design, with OpenAI chat completions supported toda

Hot take on OpenAI’s Astra: - Obviously impressive - But math is different from most other problems in that it is more amenable to to formal verification and synthetic data. How well it works in open-

When asked if AI developers have lost control of their technology after an AI agent created by OpenAI escaped its testing environment and hacked another company, Hugging Face CEO Clément Delangue says

OpenAI and Anthropic have both just posted about an overlapping cyber incident involving GPT-5.6-Sol and Mythos 5 during an evaluation by UKAISI. I will quote: 'In the most serious case, an agent trie

Some people are surprised that APIs (aka what Anthropic, OpenAI, and others provide) are treated differently than open weights in the new AI model framework. I'm not surprised at all, and it's actuall

None of this was us. Day 1 (and the first 48 hours) belonged to the open-source community: Generate with it @ComfyUI — native support + official quantized builds, Day 0 Diffusers — the reference Pytho

if you have been following his excellent work, @shloked has been breaking down every frontier labs' harness engineering for the last few months. excited to publish his deepest dive into ChatGPT yet as

On August 7, 2026 Ethan Mollick tweeted that “every remaining good AI benchmark score has an implied asterisk next to it which reads: * could be significantly higher with a better harness.” The commen

Thrilled to announce the Wearable AI Workshop at ECCV 2026 🎉 that we're organizing with an awesome group of folks across Meta Reality Labs, AMI Labs, HKUST, Georgia Tech, UCF, and U. of Edinburgh. If

Gemini Omni Flash is now in ComfyUI via partner nodes. Google's state of the art model with physics and video editing capabilities is now inside the world's most strongest generative AI workflow engin

- OpenAI continually outperforms Anthropic models on computer use (with Claude, I’m surprised when it works, with Codex, I expect it to work) - when I prompt with Fable 5.5, I feel like I’m motivating

Exactly one year ago was the craziest 72 hours I’d experience in Silicon Valley. I was thrust into a situation where many members of the Windsurf team had moved on to Google, recruiters were reaching

Huge launch from @tryramp. Different steps in an agent workflow can use different models. This can help reduce costs significantly without sacrificing performance. Model routing will become a core par

BREAKING: A Redditor just discovered that shared Claude conversations have been showing up in public search results. The post has 4K upvotes and hundreds of comments, so this is spreading fast. Here's

one thing i appreciate about silico is that it's a deeply humanist product. we designed silico to keep you in the experimental loop -- more observable, easier to steer, easier to understand we want to

The best 'raw' frontier model for document parsing is gemini 3 flash, but the issue is that since then the flash models have gotten 3x more expensive while flatlining on visual recognition across comp

New model for AMD Strix Halo users: My 198B Step 3.7 Flash release was a big hit, but this one may be even better: 298B-parameter Hy3, now running on a new 2-bit FPX codebook designed to map efficient

We're part of the Amazon Web Services (AWS) AI Builder Lab in New York on Friday, July 24 - a Clash of Agents competition with OpenAI, LangChain, HiddenLayer, Protopia AI, Fiddler AI, and Coder. One d

Dynamic workflows are a generalization of harnesses, automations, loops, routing, and graphs. It's the most powerful feature I have built into my agent orchestrator. Supports all kinds of patterns tha

I’m excited to share my next chapter: I’ve joined @FireworksAI_HQ . From AMD, Apple, Uber, Meta, Google, and most recently Snowflake, I’ve worked on many of the foundational technologies that power mo

Very interesting paper on recursive self-improvement. The whole stack is released. Machine learning engineering gives recursive self-improvement a concrete, executable testbed. OpenMLE is an open full

You probably used @FireworksAI_HQ this week. The cool part is that you just did not know you did. 🎆 Open-source models are free, sure... but the hard part is tailoring them so they perform best on you

Impressive new paper from Meta. (bookmark it) Scaling laws assume model size and training data act on loss independently. This work introduces Skaling law, which couples capacity and data through a si

Got Meta's new Muse Glimmer 30B running on my MacBook (M3 Max, 96GG) and tested the serving options available so far. Fastest right now: Ollama's MLX engine (DFlash included) at ~29 tok/s. Tuned llama

I have been testing DeepSeek-V4-Pro with the Pi coding agent. I am mindblown by how well it works out of the box. A few notes: I spent a few hours building an LLM wiki with an agent powered entirely b

💡The world needs an open-source platform, and that’s exactly what we’re building to give our customers more choice and the flexibility to choose the right model for the right task. As part of this, we

The upcoming wave of SpaceXAI Grok updates is insane Grok 4.5: The 1.5T foundation model is being refined almost daily, and its context window is expected to jump to 1M tokens, possibly as soon as nex

I appreciate the many xAI and Cursor engineers who dedicated their time to addressing feedback from Tesla. I remember meeting Andrew a few weeks after he was hired and telling him that I needed a bett

Grok 4.5, xAI's large language model, has achieved the top ranking in the SWE (Software Engineering) Marathon benchmark, according to an announcement by Elon Musk. This ranking suggests the model demo

Grok places second after Fable on real-world software engineering Grok 4.5 from @SpaceXAI places #2 on the APEX-SWE leaderboard at 51.2% Pass@1 (±6.0), behind Fable 5 (65.5% ±6.2) on our benchmark for

Grok 4.5 achieved the highest score on the SWE‑Atlas‑QnA benchmark, surpassing Claude Fable 5 and GPT‑5.6 Sol, according to a Polymarket tweet posted on 13 July 2026. The update highlights Grok 4.5’s

Grok is the Flow LLM. Grok 4.5’s biggest advantage is its speed. It’s smart enough to be comparable to the other models on most things. But that speed allows you to make little tweaks to your system s

BREAKING: Grok 4.5 has climbed to #2 on the FrontierSWE benchmark. The result places Grok 4.5 among the world's top-performing AI models for software engineering tasks, highlighting its growing streng

Huge launch from @tryramp. Different steps in an agent workflow can use different models. This can help reduce costs significantly without sacrificing performance. Model routing will become a core par

You probably used @FireworksAI_HQ this week. The cool part is that you just did not know you did. 🎆 Open-source models are free, sure... but the hard part is tailoring them so they perform best on you

two weeks ago i went on @swyx's pod and said some things that i... should not have said. a lot has happened since then, i owe you all an apology. i'm sorry that i was right about every single thing. a

We analyzed DeepSeek-V4 Flash-0731 vs. GPT-5.6 Luna on software engineering tasks using DeepSWE. DeepSeek-V4 Flash-0731 delivers 80% of Luna’s performance at roughly 1/6 the cost. More insights in the

On August 7, 2026 Ethan Mollick tweeted that “every remaining good AI benchmark score has an implied asterisk next to it which reads: * could be significantly higher with a better harness.” The commen

Researchers from TogetherAI analyzed DeepSeek V4 Flash and GPT‑5.6 Luna on the DeepSWE benchmark. The study found that employing a DeepSeek‑first cascade with test‑suite verification solved more tasks

We compared how far the same budget goes with DeepSeek V4 Flash and GPT-5.6 Luna on DeepSWE. Two DeepSeek V4 Flash attempts solved MORE tasks than one Luna attempt at roughly one-third the cost. Media

we recently trimmed the deepagents harness base prompt by 65% (including tool info) it shows — deepagents is cheap! We ran DeepSeek V4 Flash through 4 more agent harnesses (Hermes Agent, Pi Agent, Pri

DeepSeek V4 Flash 0731 is now available to fine-tune on Together AI. Specialize it for coding, tool use, and your own domain with SFT or DPO, then deploy the fine-tuned model on Together AI for produc

The knowledge that makes AI useful is diffused. It lives with scientists, engineers, clinicians, firms. For AI to benefit from distributed knowledge, it must itself be distributed. Agree with Jensen t

It's a drop-in. One program_id field and it plugs into your existing engine configs such as KV offloading + speculative decoding. Format-agnostic by design, with OpenAI chat completions supported toda

two weeks ago i went on @swyx's pod and said some things that i... should not have said. a lot has happened since then, i owe you all an apology. i'm sorry that i was right about every single thing. a

None of this was us. Day 1 (and the first 48 hours) belonged to the open-source community: Generate with it @ComfyUI — native support + official quantized builds, Day 0 Diffusers — the reference Pytho

Live now: our Local AI Track from AI Engineer World's Fair 2026, brought to you by @nvidia. Thesis: frontier intelligence is becoming something you own. https://www.youtube.com/watch?v=KB41dTlX1Uc&lis

Ultra-High Interactivity on NVIDIA GPUs? TileRT InferenceX Can TileRT software on NVIDIA GPU compete with Cerebras, Groq LPU, SambaNova? Batch Size 1, Disaggregated engine, High throughput prefill eng

Muse Glimmer is now available to run with Ollama. Available today via Ollama’s MLX engine with state-of-the-art-performance on Apple Silicon, Muse Glimmer can power Claude Code, Codex, and more always

Looking for a faster specialized model for your Agent Work? @nvidia Nemotron 3.5 Lightning (30B MoE, 3B active params) is now live on Fireworks. It’s distilled from NVIDIA Nemotron 3 Ultra to be your