- AI Fire
- Posts
- 🔬 Scientists Just Got Outsmarted by AI
Read time: 5 minutes
Google’s new Stitch 2.0 could be the free AI agent that flips coding + design forever. YouTube rolled out its most ambitious AI update yet. And buried in a new study? AI just proved it can out-invent scientists themselves.
What are on FIRE 🔥
IN PARTNERSHIP WITH THE MARKETING MILLENNIALS
Marketing ideas for marketers who hate boring
The best marketing ideas come from marketers who live it. That’s what The Marketing Millennials delivers: real insights, fresh takes, and no fluff. Written by Daniel Murray, a marketer who knows what works, this newsletter cuts through the noise so you can stop guessing and start winning. Subscribe and level up your marketing game.
AI INSIGHTS
A new paper (Sep 2025) shows how an AI + Tree Search system can now write expert-level scientific software - and in many cases, beat the top human-designed methods.
Key Results:
Bioinformatics: Created 40 new methods for single-cell RNA analysis that climbed past leading tools on the public leaderboard.
Epidemiology: Built 14 models that outperformed the CDC’s official ensemble and every single human-designed COVID-19 forecast model.
Other domains: Hit state-of-the-art in:
Geospatial analysis for land cover + deforestation studies (used by NASA & Google Earth data)
Neural activity prediction in zebrafish (a benchmark from top neuroscience labs)
Time-series forecasting on the GiftEval benchmark (a gold-standard used in academic ML research)
Solving symbolic math integrals with SymPy and LLM pipelines (a tool trusted by millions of engineers worldwide)
Benchmarks:
Validated across 6 different scientific fields with strict peer benchmarks.
In single-cell RNA tasks, the AI’s Batch Balanced KNN (TS) beat the best method (ComBat, a widely used open-source tool) by 14% and outperformed in 11 of 13 metrics.
Instead of just tuning like AutoML or Copilot, it invented new algorithms by recombining existing ones (e.g., fusing ComBat + BBKNN into something better).
Why it matters: Just like GitHub Copilot changed how developers code, this shows AI can now out-invent scientists in core methods. With trusted benchmarks from the CDC, SymPy, and leading bioinformatics platforms, this isn’t hype - it’s a glimpse of research labs where AI doesn’t just support scientists, it competes with them.
PRESENTED BY I HATE IT HERE
HR is lonely. It doesn’t have to be.
The best HR advice comes from those in the trenches. That’s what this is: real-world HR insights delivered in a newsletter from Hebba Youssef, a Chief People Officer who’s been there. Practical, real strategies with a dash of humor. Because HR shouldn’t be thankless—and you shouldn’t be alone in it.
TODAY IN AI
AI HIGHLIGHTS
🎬 YouTube just packed creator mode: inspiration tab, title A/B tests, auto-dubs, “likeness” detection (open beta), AI highlights for Live, and a Veo 3 Fast upgrade for Shorts. This all dropped at Made on YouTube - plus new brand-deal tools and Shopping tags for easier monetization.
🧭 Bloomberg Weekend argues the ladder isn’t broken: junior roles will adapt faster, usage data backs it, and history rhymes (PC era all over again). Read the case for AI Won’t End Entry-Level Work - with fresh stats from Pew, Anthropic, and OpenAI.
🕶️ William Blair came out bullish after Meta Connect 2025: new Ray-Ban/Oakley smart glasses lineup, better video + battery, broader AI wearable use cases. They still see long-term AI leadership.
☁️ Goldman Sachs raised its target on Baidu: Cloud + AI are the growth engines; non-search could hit 54% of core revenue by 2027. Apollo Robotaxi, chip-to-software stack, and shareholder returns round out the thesis.
💰 AI Daily Fundraising: Terra Security raised $30M in Series A, led by Felicis with backing from Dell Technologies Capital, SVCI, SYN Ventures, LAMA Partners, and Underscore VC. The company, which pioneered agentic-AI powered continuous penetration testing, has now secured $38M in total funding.
AI SOURCES FROM AI FIRE
NEW EMPOWERED AI TOOLS
📋 Dart is the only AI-native project management tool, with chat + agents to plan and execute faster
📽️ Animant 2.0 turns slides into immersive 3D presentations with audio, video, and interactive models
📊 Teable makes your data actionable, from invoices to customer insights and marketing content
🧠 Grok 4 Fast (by xAI) is now live, the multimodal AI model with a 2M context window and top efficiency
AI QUICK HITS
🌍 Dell says Israeli innovation is quietly powering its AI push - Meier teased PowerFlex Ultra, AI PCs with NPU teraflops, and measurable sales lift without higher costs.
🚗 Tesla got upgraded to Outperform at Baird on a coming physical AI era - Optimus/robotaxis on deck, and Musk’s mega-milestone plan in focus.
🧨 Nvidia’s Jensen Huang is “disappointed” amid a reported China chip ban on RTX Pro 6000D - adding fresh strain to already volatile U.S.–China AI trade.
🧰 Startup ops hack: a crisp Top 10 Project Manager Assistant AI list (ClickUp, Asana, Wrike, Motion, Notion AI, and more) to cut coordination drag and ship faster.
🛠️ Zapier’s HR Automation Engineer shared her favorite Zapier Agents - Slack to-do capturer, morning digest, and LinkedIn balancer- with plug-and-play templates.
AI CHART
Mistral AI just dropped Magistral Small 1.2 and Magistral Medium 1.2, upgrades to their 1.1 models - and the boost is clear.
Multimodal now → Both models gained a vision encoder, so they handle text + images in one go.
Performance jump → +15% on math/coding benchmarks like AIME 24/25 and LiveCodeBench v5/v6.
Smarter tools → Better at web search, code interpreter, and even image generation.
Cleaner output → More natural tone, sharper formatting, fewer infinite loops.
For context:
Magistral Small 1.2 fits on a single RTX 4090 or MacBook with 32GB RAM when quantized.
Magistral Medium 1.2 is already hitting 91.8% pass@1 on AIME24, a massive leap from 72% in v1.1.
Mistral is signaling something big: these aren’t just efficiency tweaks - they’re reasoning-ready models that can run locally, at scale, and compete with anything from OpenAI or Anthropic.
We read your emails, comments, and poll replies daily
How would you rate today’s newsletter?Your feedback helps us create the best newsletter possible |
Hit reply and say Hello – we'd love to hear from you!
Like what you're reading? Forward it to friends, and they can sign up here.
Cheers,
The AI Fire Team
Reply