Search: “agents”

ml-intern is fully on mobile now you can launch 8 A100s from your phone. while on the couch. while commuting. wherever I just did this while biking. same sessions as your desktop too — start a run on

// When to Retrieve During Reasoning // Pay attention to this one, AI devs. (bookmark it) Most RAG systems retrieve once, before the model starts reasoning. Large reasoning models like o1 and R1 don't

A million-token context window is not a strategy. 🛑 Our Head of DevRel, @RoieSchwabco , explains why dumping data is killing your RAG performance: 📍 One needle in a haystack? Easy. 📍 Multiple needles?

IMO DeepSeek v4 demonstrated utter confidence and competence by not benchmaxxing, not focusing on some BS final run cost, not even spending inference-optimal compute. just showed up, demonstrated SOTA

Recreated the app with ml intern to compare. Went faster than in cowork, some things are better (ex direct integration in the desktop reachy mini app), some things are worse (had to debut the install)

Excited to support @NVIDIA Nemotron 3 Nano Omni, now available on Fireworks. It's the first open model that handles vision, audio, video, and text in a single inference loop. Built for multimodal sub-

First open-weight model from @poolsideai! Apache license, and available on Ollama to try. 👇👇👇 model page Today we’re releasing Laguna XS.2, Poolside’s first open-weight model. It’s a 33B total / 3B ac

GLM 5.1 from @Zai_org is now available on @FireworksAI_HQ Training Platform across the Managed and Training API workflows. Try SFT and DPO with smart defaults or your own custom loss function with a 2

No matter what kind of company you are...start making your internal company data legible to AI. Today. As a founder, you are essentially building two versions of your company: the one humans work in a

@NVIDIA Nemotron 3 Nano Omni is now on Together AI. Enterprise multimodal AI — video, audio, image, documents & text — optimized for speed and scale. ✅ ~3B active params, 9x higher throughput ✅ Fully

Open Models & a potential looming Haiku-apocalypse? 🚀 chart comparison UI courtesy of @OpenRouter there's a large class of problems where you don't need frontier intelligence. you want cheap, fast, an

Want to see which frontier models do the best on document understanding? Check out our ParseBench leaderboard on @kaggle! https://www.kaggle.com/benchmarks/llamaindex-org/parsebench For more details o

“Always keep a separate backup.” For sure, every experienced programmer should know this. 𝘉𝘶𝘵 𝘸𝘦 𝘫𝘶𝘴𝘵 𝘤𝘳𝘦𝘢𝘵𝘦𝘥 𝘢 𝘸𝘩𝘰𝘭𝘦 𝘨𝘦𝘯𝘦𝘳𝘢𝘵𝘪𝘰𝘯 𝘰𝘧 𝘷𝘪𝘣𝘦 𝘤𝘰𝘥𝘦𝘳𝘴 𝘸𝘩𝘰 𝘥𝘰𝘯’𝘵. And we now have a legion of synthetic coding

going to buy 2 rtx 6000 just because of the capabilities of local models becoming great! no more outsourcing of research to the api Xiaomi MiMo-V2.5 is now officially open-sourced！ MIT License, suppor

I'm HIRING 5 more people to help me build the future of AI media. The Rundown is closing in on 3M active readers, and the demand is far outpacing what we can handle. 2,000 referral bonus if you help u

Loan processors spend 40–60% of their time reconciling income across tax returns, pay stubs, W-2s, and bank statements. We built an end-to-end pipeline that automates it with LlamaParse + the Claude A

People keep blaming the users for the growing number of vibe-coded reasoners. Doing so is *half* right. Users *are*screwing up - by letting vibe coded stuff access their files, without proper backups,

.@swyx on whether AI infrastructure has finally stabilized: Media Always enjoy getting to chat with @swyx on our annual cross-episode with @latentspacepod on the state of AI. We hit on what’s shifted,

Continued weak spots of AI, from the point of view of a business professional and not a PhD biochemist: 1) SVGs. The ability to 'illustrate' and have that thing be infinitely scalable. See below image

I am increasingly bullish on open source harnesses (like OpenCode) not because they will be better than SOTA closed harnesses, but because they will never pull shady stuff like what Claude Code and ot

Opened a llama.cpp discussion about whether custom GBNF grammars can compose with tool calls in llama-server. Right now tools work alone, grammar works alone, but tools+grammar doesn't. If you use lla

The community can now download pre-quantized weights from MLX community repo on HF thanks to @LambdaAPI Model collection: https://huggingface.co/collections/mlx-community/deepseek-v4 DeepSeek-V4-Flash

Launching pyptx — a Python DSL for writing NVIDIA PTX kernels. One PTX instruction = one Python call. Write pure PTX in Python. Direct Hopper + Blackwell support: wgmma, TMA, tcgen05, mbarriers. JAX +

Local models do seem likely to create an explosion of new use cases. Local compute >> cloud compute. This is where we are right now. And i’m not gonna lie it feels pretty magical 🧚‍♀️ Qwen3.6 27B runn

Fireworks AI posted a question to their audience asking what features or capabilities they would like to see developed next, with a specific mention of full parameter tuning as a potential option. Thi

4⃣4⃣4⃣4⃣ Introducing DeepSeek V4 Pro, a long-context model with hybrid attention, three reasoning modes, and SOTA coding performance. AI natives can now use DeepSeek V4 Pro on Together AI and benefit

Cloud-hosted models will become more niche/seldom-used as much cheaper local models fulfill that which 90% of consumers need from AI. Lots of models to buy/use, but very few hardware options upon whic

Congrats to @deepseek_ai team! Doing the numbers I would estimate: Pro < 14m for the final training run Flash < 4m Ratio of active params x total training tokens vs v3 Total compute costs (data prep,

For the last 72 hours since ml-intern launched we have had over 500+ autonomous AI research projects running on the Space at all times. Some insane ones I saw: 1. A new AI paradigm from scratch — tryi

Introducing DeepSeek V4 Pro, a long-context model with hybrid attention, three reasoning modes, and SOTA coding performance. AI natives can now use DeepSeek V4 Pro on Together AI and benefit from reli

Just merged a built-in skill for Google's DESIGN.md A skill that lets Hermes author, lint, diff, and export DESIGN.md files, giving it fluency in Google's new open-source visual-identity format the mo

Models are changing really fast, often in ways that will break your product and technical assumptions. There's no way around this – the only way forward is to relentlessly iterate to make sure you're

CursorBench is a benchmarking tool or evaluation framework developed by Cursor, likely designed to measure and compare the performance of AI code completion or development features. The post provides

New episode of The Information Bottleneck is out, this time with @liuzhuang1234 (Princeton). We talked about ConvNeXt and whether architecture still matters; dataset bias and what 'good data' actually

This has been the most anticipated event in OSS and it doesn't disappoint. Congratulations to @deepseek_ai team. DSV4 Pro is now available on @togethercompute and we'll be adding a lot of capacity beh

At #RenderCon2026, @EMostaque makes the case during How to Build the Holodeck that stories still matter, IP still matters, but the barriers to building entirely new experiences are dropping faster tha

GPT-5.5 in Codex is a delight to work with: - Super sharp with responses - It understands intent better than any model - Great 'personality' - Gets lots of stuff done without pausing unnecessarily It

I've been previewing this in Codex for a few weeks - it's very good! Had some great results from it having it run security reviews against code written using other models Introducing GPT-5.5 A new cla

looks like new Pareto frontiers across everything: - Context: 400K context in Codex and a 1M in API - API Pricing: 5/m input and 30/m output tokens. - Codex improved its own inference speed 20% lol -

🚨 OpenAI just launched GPT-5.5. The OpenAI team was nice enough to give me early access over the last several weeks, and I just want to flag: there is a certain class of models (one that we’re hitting

Things have been degrading super fast in Claude Code. I still use Claude Code, but my default is now Codex. I still prefer Opus models for coding, and so I will try again with the fixes. I appreciate

Boris Cherny announced that usage limits are being reset for subscribers in response to user feedback and patience. The post suggests this is a policy adjustment made by a service or platform to addre

This is a really big deal - it's easy to run into nasty bills with Cloud Run if your site attracts aggressive scrapers, spending caps make it a whole lot safer to run small projects on Anouncing Spend

I built a web analytics system to track traffic and events across all my @Replit projects in one place. Drop-in @posthog stack: → Full analytics dashboard → Reverse proxy (beat ad blockers) → AI skill

I find that open weights models over-perform on benchmarks compared to actual real-world usage, and Kimi feels like no exception. For example, a small amount of use will show that Kimi is not as good

K2.6 + hermes = 4 hr session setting up qwen 3.6 training regime on dgx spark with the current autnomous session currently lasting 70+ min without any prompting. Kimi with hermes is next level. @NousR