AI with Kyle Daily Update 046

Today in AI: Mistral Raises $1.3B, OpenAI Admits Benchmark Problem, AI Jobs Safe at Top

Author

Kyle Balmer
September 10, 2025

Streaming live daily at 10AM UK time on YouTube (and all other platforms).
Catch me here: Subscribe to ‘AI with Kyle’ Channel

The skinny on what's happening in AI - straight from the previous live session:


Highlights

🇪🇺 Mistral Raises €1.3B But Remains 10x Behind US Players

European AI champion Mistral secured €1.3 billion from Dutch chip equipment maker ASML, bringing their total funding to €1.7 billion and valuation to nearly $12 billion.

Never heard of Mistral? Not that surprising. Mistral currently sits 14th on LM Arena leaderboards - competitive but not front-running. The funding comes as the EU considers building its own large language model.

Kyle's take: This is brilliant for European AI, but let's be realistic about the scale problem. Anthropic raised $13 billion last week - ten times this amount. (OK sure they immediately lost 10% to the author copyright settlement…)

The current AI paradigm is basically "throw more compute at the problem," which means you need massive data centres and training budgets running into billions. €1.3 billion is huge money, but it's probably not enough to catch up with the Americans and Chinese who are already ahead.

Unless Mistral finds a fundamentally better way to train models that doesn't require as much compute, they're fighting an uphill battle with a smaller war chest.

Even so: félicitations!

💼 OpenAI Hires £300K Content Strategist - AI Jobs Paradox Explained

OpenAI posted a content strategist role paying $300,000-$393,000 plus equity, sparking criticism about hiring humans for jobs AI should handle. Critics of AI are using this listing to score cheap points.

Kyle's take: People are missing the nuance here. AI isn't sweeping across every job at the same pace - it's following a "jagged edge" pattern, hitting different roles at different speeds.

A $300K content strategist working for a multi-billion-dollar company is at the top of their game. That's completely different from entry-level copywriting on Fiverr, which AI is decimating. We're seeing this pattern everywhere: junior roles down 23%, senior roles up 14%. It’s a false equivalence to think of all “content strategists” as the same - OR any job. There are gradations within industries and roles.

🧠 OpenAI Admits AI Benchmarks Are Making Models Lie More

OpenAI published research showing AI models hallucinate because they're overtrained on benchmarks that penalise saying "I don't know." Models learn to give plausible-sounding answers rather than admitting uncertainty because test scoring systems give zero points for honest ignorance. The paper calls for changing how we evaluate AI systems.

Kyle's take: This is refreshing honesty from OpenAI - admitting their own testing methods are part of the problem! Everyone is chasing the benchmarks because it makes them look better - even if it makes the model less useful in real life.

It's like students cramming for exams who learn to game the test rather than actually understanding the subject. When GPT-5 occasionally says "I don't know" instead of making something up, that's actually brilliant progress.

We need AI systems that admit uncertainty rather than confidently spouting rubbish. The fact that OpenAI is willing to call out the entire industry's benchmark obsession shows they're serious about building more trustworthy systems.

Now let’s see if it changes anything!

Source:

📊 UK Poll: 73% Want to Halt AI Development (But Questions Are Dodgy)

Control AI published a YouGov poll of 2,344 UK adults showing 73% favour halting "superintelligence" development and 82% want AI safety testing.

Kyle's take: I'm suspicious of these numbers. If you ask random people whether they support "prohibiting superintelligent AI systems smarter than humans," of course they'll say yes - it sounds terrifying and makes people think of Terminator.

But do they actually know what superintelligence means? This feels like asking about flying cars or time travel. The anti-AI groups behind this (Control AI, Pause AI etc.) have tiny followings at the moment - 16,000 Twitter followers isn't exactly a mass movement. It doesn’t feel like the public really cares that much…

That said, fear of new technology is completely normal!

Teachers literally went on strike to protest calculators being introduced in schools. Every disruptive technology gets this pushback. We need to have these discussions but leading questions in a survey is still a little tricksy!

Member Question: "What fundamentals do I need to start my journey to LLMs and AI mastery?"

Kyle's response: Depends what level you want to reach! For daily usage, just start using AI tools - that's honestly the best way to learn. For building AI products, check out Lovable or Cursor and learn to create software with AI assistance. For understanding how models actually work, start with 3Brown1Blue’s videos on neural networks - brilliant explanations without getting too mathematical.

Then look at Andrej Karpathy's longer YouTube videos and courses on deeplearning.ai. But here's the key: you don't need to understand the deep mathematics and tech to use these tools effectively, just like you don't need to understand HTTP protocols to browse the web.

📚 Free Resource: Philosophy of AI 10-Week Course

Dan Williams from Conspicuous Cognition has released a comprehensive 10-week syllabus covering the philosophy of artificial intelligence. The course includes readings, videos, and covers everything from "Can computers think?" to "Will AI kill us all?" Each week has structured materials and from quick scan it’s all great stuff.

This looks fantastic - I'm considering doing it as a live group study. The content is serious academic material, so only for people really committed to understanding AI philosophy, not just using the tools. Not required for day to day use but extremely interesting!

Find it here: Conspicuous Cognition. 

Want the full unfiltered discussion? Join me tomorrow for the daily AI news live stream where we dig into the stories and you can ask questions directly.

Streaming on YouTube (with screen share) and TikTok (follow and turn on notifications for Live Notification).

Audio Podcast on iTunes and Spotify.