All

Updated list of open LLMs you can use commercially. Handy when picking a model for a client project.

Ollama on a laptop is great for tinkering, but once you need reliability, hosted APIs win on latency and uptime. My experience so far.

Every framework becomes a debugging nightmare past 500 lines. Plain code + function calling has been better for us.

All data stays on device. Weekly digest with topics, mood, and action items. 400 users in 2 weeks!

Multiple model sizes, powering products across Meta. The open-vs-closed debate keeps heating up.

MathBench leaderboard flipped again. The open-source gap is closing faster than expected.

Doctors' AI scribe just raised a huge round at $5.3B valuation. Healthcare AI is the real money-maker apparently.

Ticket deflection went from 0% to 38% in a month. Here's the architecture and the mistakes we made.

We went LibreChat for multi-user + RBAC. Open-webui has nicer UX though. Your pick?

Mine: Claude for code, DeepSeek for bulk tasks, local Ollama for privacy stuff. What's yours?

We use RAGAS but the scores don't correlate with human judgment. What do you actually use in production?

Scrapes 20 sources, dedupes, summarizes, posts at 8am. 1.2k subscribers now. AMA about the pipeline.

Big claim. If real, this changes agent UX a lot. Anyone benchmarked it yet?

Keep this tab open. Everything from o1 to GPT-5-Codex-Mini lands here first.

34 of the 100 fastest-growing companies are AI. The capital shift is real and it's accelerating.

Chroma vs Qdrant vs LanceDB for a small self-hosted RAG service. Disk and RAM matter — running on a free-tier VM.

Ran a quick benchmark. pgvector with HNSW was 80% the speed at 10% the cost. Details in thread.

Both raised big rounds in 2025. If you're building on top of LLMs, infra plays look safer than apps.

Feed it a CSV export, it finds anomalies and writes a Korean explanation via an LLM. Useful for MSP folks.

Prefill pipelining finally landed. Our 128k-context serving costs should drop meaningfully.

One endpoint for all providers, consistent interface, and the config is plain YAML. We swapped out three vendor SDKs in an afternoon. If you're multi-provider, just use it.

Long agent sessions blow up tokens. Do you use summarization, sliding windows, or something else?

I packaged my go-to RAG setup into a template. Docker compose up and you have a working API in 2 minutes.

Replaced a custom fine-tune with a generic 8B + good retrieval. Cheaper to maintain, better scores. Data point for you.

Worth a read — some genuinely interesting verticals (last-mile routing, predictive maintenance).

Woke up to the leaderboard shuffle. Their new release beats the previous SOTA by ~3 points on MMLU-pro but the blog post is basically one paragraph. Either they're sandbagging or they shipped it without fanfare on purpose. Either way the pricing page is still the old one, which is the real story.

It joins the call with a normal name, takes notes, and posts them to our wiki. People forget it's there until the notes appear. Uses a tiny local model for transcription. The 'is this legal' question came up once. HR said fine, so we're shipping it.

Model cites chunks that don't actually support the answer. Any post-processing tricks that work?

vLLM vs TGI vs llama.cpp server. For a small team serving one 8B model, what's the least ops burden?

Visual node editor for agent pipelines with live tracing. MIT licensed, would love feedback.

CSV of prompts → JSONL of responses + scores. 100 lines of Go. Maybe useful for your eval pipeline.

Gradio got multimodal chat components. Streamlit is catching up. For internal demos, which wins?

With better rerankers and hybrid search, raw embedding quality matters less every quarter. Agree?

I used to read every arxiv paper. Now I wait for good summaries. Am I falling behind?

Small enough to self-host, strong enough for real agent loops. Worth a look for cost-sensitive teams.

Scrapes arXiv + HN, summarizes with an LLM, posts as threads. 200 stars in a week. AMA.

I keep coming back to plain code + function calling. Frameworks add too much magic for debugging.

Lots of 'prompt engineer' listings that are really content farms. Anyone found legit remote AI work lately?

Downloaded the 27B variant last night. Quantized to Q4 it fits in 16GB RAM and generates at a very usable speed. Not Claude-level reasoning, but for a local model it's a big step. Google is winning the 'open weights you can actually run' game right now.

This week alone: three new models, two frameworks, one protocol update. I've started ignoring anything that isn't usable in my stack today. My feed is a graveyard of 'this changes everything' posts I never read. What's your filter?

Ship a model, top the leaderboard, get the press release, repeat. Meanwhile real usage is full of edge cases none of the benchmarks touch. I'd rather have a model that's honest about what it can't do.

Our org has 10k pages of policy PDFs nobody reads. Built a RAG bot with citations that links back to the exact page. Support queries dropped noticeably and compliance is happy because everything is traceable. Side project that became the team's favorite tool.

System prompt first, stable ordering, minimal drift. Any other tricks to keep cache hits high?

We need to ground answers in internal docs that update weekly. RAG feels obvious but retrieval quality is killing us. Fine-tuning is a one-time cost but the docs change. Anyone run a hybrid? What's your split?

We were merging PRs blind. Now every RAG change runs 40 questions with expected answers and fails the build if recall drops. It's crude but it caught two regressions last week. Sometimes the simple stuff wins.

They released the evals and methodology this time, which is rare. HumanEval pass@1 is a couple points ahead of Llama 4.5 and it's faster to serve. We're spinning up a test pod this week. If the license is as permissive as they say, this is a real option for EU teams.

Just got the email. Cache writes are 50% cheaper and cache reads dropped too. For anyone running long agent loops this is basically free money. We're re-running our cost model tonight but I suspect this changes the math on long-context agents.

Long context is expensive and my tests show the model forgets the middle anyway. Chunking into a RAG loop works but feels like giving up. Do you just pay for big context or actually engineer around it?

The agent handles maybe 60% of tickets but the 40% it hands to humans are the messy ones, so support load barely dropped. Customers who realize they're talking to a bot get annoyed even when it works. Not sure this was a win.

50 lines of TypeScript, points at our search API, and now everyone queries docs from their editor. It's dumb but it changed how we work. The protocol is clunky but the integration win is real.