Two big data drops today, and they tell the same story from different angles.
Anthropic's 400K session study confirms what we suspected: domain expertise now beats coding skill. Lawyers and managers succeed at nearly the same rate as engineers. Meanwhile, OpenAI trained GPT-5.5 Instant with 600 doctors to cut health hallucinations 71%. Doctors as training signal, not just end users.
The pattern is clear: AI is absorbing specialist knowledge and redistributing it. Your moat is shrinking.
|
|
In Partnership with
|
·
|
|
|
|
|
|
AlphaSignal Webinar
|
There's a name for what's happening when your AI coding tools ship code that looks right but fails in production: the Context Void. It's the gap between what your agent knows and what it actually needs to know to write reliable code. Most teams don't realize they're falling into it until something breaks. On June 25th at 10am PT, AlphaSignal is sitting down with Gal Vered, CEO of Checksum, to break down exactly how the Context Void works and what it takes to close it. Join us for this free webinar to understand why your AI-generated code keeps failing in production and what to do about it.
|
|
|
|
Top Repo
|
|
|
|
Unsloth shrinks GLM-5.2 by 84% so it runs locally on a 256GB Mac
|
|
5,769 Likes
|
|
A 1.51TB AI model running on your laptop sounds impossible. It kind of is. But Unsloth just made it a lot less impossible. GLM-5.2 is Z.ai's newest open model, and it's a big one: 744 billion parameters, MIT licensed, with a 1M token context window (think: entire codebases at once). The catch? The full model weighs 1.51TB. Nobody has that lying around. Unsloth's fix: quantization. That means shrinking the model by storing numbers less precisely, like compressing a photo. The 2-bit version drops to 238GB, retaining ~82% accuracy. Important layers stay at higher precision so quality doesn't collapse. What you actually need to run it: - A 256GB Mac (M3/M4 Ultra) for single-box setup
- Or a 24GB GPU plus 256GB RAM with memory offloading
- Tools: llama.cpp, LM Studio, or Unsloth Studio
This opens up private, offline coding agents and long-context tasks that previously required a cloud API. No usage fees, no rate limits, no data leaving your machine.
|
|
|
|
Presented by Unblocked
|
|
Your agents are still missing context.
|
|
AI shows up in 60% of engineering work. But only about a fifth of it can be handed off without someone babysitting the output. That’s because agents are missing context. This 8-level context maturity model shows why your output is still inconsistent after all the tokens burned. Join Unblocked live June 24 (FREE) to learn: - Why more MCPs provides agents access but not understanding
- What it takes to deploy agents you can trust without supervision
- How a context layer solves for quality, efficiency and cost
|
|
|
|
Top Repo
|
|
|
|
AgentDeck plugin turns Stream Deck+ into a Claude Code control panel
|
|
5,125 Likes
|
|
Tired of tab-switching every time Claude Code needs your input? Someone just turned a Stream Deck+ into a physical control panel for AI coding sessions, and it's exactly as cool as it sounds. The setup uses AgentDeck, an open-source plugin that hooks into Claude Code's event system. Claude Code fires off signals whenever something happens in your session. AgentDeck catches those signals and pushes live status updates to the physical buttons and dials on your Stream Deck+. What you can actually do with it: - See which sessions are running, waiting for input, idle, or errored, all on physical buttons
- Approve or deny permission requests with a single button press, no keyboard needed
- Monitor token usage and session cost per button
- Manage multiple Claude Code sessions simultaneously from one surface
To try it: grab the repo at github.com/asheshgoplani/agent-deck and run the install script. Currently Stream Deck+ only.
|
|
|
|
Presented by Datalab
|
|
Docs to records that fit your schema. Open source.
|
|
Datalab just released Lift, an open source structured extraction model: - Pulls user-defined fields out into your schema
- Any schema - invoices, scientific papers, 10-Ks, you name it
- Fast - 9.5s median, 8x quicker than Azure
- Run locally or via Datalab’s API for scale
Schema in, structured values out.
|
|
|
|
Top News
|
|
|
|
Anthropic study of 400K Claude Code sessions shows task value grew 27% in six months
|
|
5,081 Likes
|
|
Anthropic just dropped a study on 400,000 Claude Code sessions from 235,000 users, and the findings are worth knowing. Here's what the data actually shows: - You don't need to be a software engineer. Every major occupation succeeded at coding tasks within 7 percentage points of software engineers. Lawyers, finance folks, managers -- all roughly the same success rate.
- Sessions are getting more valuable. Compared against freelance marketplace rates, the average session grew 27% in monetary value from October to April. People are tackling harder, pricier work.
- The work split is clear. You handle ~70% of planning decisions. Claude handles ~80% of execution.
- Expertise still matters. Novices hit verified success 15% of the time. Intermediate and expert users land between 28-33%. The gap between intermediate and expert is small though -- being decent at a domain is enough.
- Use cases are shifting. Code-fixing sessions dropped from 33% to 19%, while software operation grew from 14% to 21%.
The big takeaway: Claude Code is becoming a tool for anyone with domain knowledge, not just people who can code.
|
|
|
At Alpha Signal, our mission is to build a sharp, engaged community focused on AI, machine learning, and cutting-edge language models, helping over 200,000 developers stay informed and ahead. We're passionate about curating the best in AI, from top research and trending technical blogs to expert insights and tailored job opportunities. We keep you connected to the breakthroughs and discussions that matter, so you can stay in the loop without endless searching. We also work closely with partners who value the future of AI, including employers and advertisers who want to reach an audience as passionate about AI as we are.
Our partnerships are based on shared values of ethics, responsibility, and a commitment to building a better world through technology.Privacy is a priority at Alpha Signal. Our Privacy Policy clearly explains how we collect, store, and use your personal and non-personal information. By using our website, you accept these terms, which you can review on our website. This policy applies across all Alpha Signal pages, outlining your rights and how to contact us if you want to adjust the use of your information. We're based in the United States. By using our site, you agree to be governed by U.S. laws.
|
|
|
Looking to promote your company, product, service, or event to 250,000+ AI developers?
|
|
|
|
|