🔥The 9th Round of Easy Loan, Earn $40 Reward is in progress❗️ ⏰ Promotion Period: January 15th - Feburary 15th, 2025 👉 Register now and check more details at gate.io/campaigns/358

i read the AlphaEvolve paper over the weekend. my analysis below! key question: how important is it? there are two dominant architectural substrates of frontier AI systems: neural networks (NN) and symbolic programs (SP). they exist at opposite ends of a spectrum with very

27% of @Remote’s IT tickets are solved automatically. Using an AI-powered IT help desk built in Zapier. (It’s saving the team $500k a year in hiring costs 👀)

ARC v2 paper is officially out! we tested v2 in a controlled setting with over 400 humans, this report contains details and analysis to substantiate our relative "easy for humans, hard for AI" claim. we'll be releasing the raw data later this week.

The OpenAI Responses API now supports tens-of-thousands of third-party actions via Zapier MCP. Demo in thread below!

Here's a paper with very interesting research on the security of LLM text embeddings. Authors claim they can reverse embeddings to text with no access to the original model and only a few ~thousand samples arxiv.org/abs/2505.12540

Interactive benchmarks are coming back 👾

And now available via Anthropic API -- today Anthropic announced the Zapier MCP connector support: anthropic.com/news/agent-cap…

Shoutouts from Anthropic and Microsoft at the same time. Zapier MCP everywhere.

IMO this is broadly a good take. I'd soften "expert" as I think it will be difficult to deny AGI once AI can do all tasks humans find relatively easy. Today it remains easy to find such tasks.

I've been looking forward to hearing about John's work for years. He was one of first high profile people to bet on alternative ideas to scaling up LLMs. New ideas still needed for AGI. Highly recommend checking this out.

AI Agent UX is converging towards email.

Claude Sonnet 4 ARC results. One broad useful note for devs is the synchronous API test-time compute limits of most TTC systems is ~linear on accuracy vs cost. Providers have picked their sync TTC points well, the exponential cost regime requires streaming API.

Please help if you can!

Most founders think great SEO is about hygiene: title tags, keywords, back links, etc. These are only table stakes to get and stay indexed. Great SEO is actually more like user research. In fact, Google is basically an online RL algorithm doing user research. Users have a broad

Thanks to friends at Anthropic we've now got Claude 4 Opus results on ARC! The most interesting finding is semi-private v2 scoring ~9% (new commercial SOTA) while v1 only at 35%.

We’re setting a new standard at Zapier. 100% of new hires must be fluent in AI.

The field of program synthesis is growing. Eventually, every AI system will leverage this paradigm.

True for most automation scenarios. Given machine intelligence on par with human capability for a task, machines offer lower variability and will be preferred (even commanding a premium on price).

Hard thing about "AI SEO" is the O. There is no feedback loop to optimize, unlike classic search which has indexing and bounce rate loops. For niche/long-tail/novel content, AI falls back to classic search anyway. For now you're better off using AI to win classic SEO.

Why wait until 2027 to release Claude pioneers?