Why data is the biggest AI bottleneck (feat. Arthur Mensch of Mistral AI) | E2212
This Week in Startups
Jason Calacanis
4.4 • 1.3K Ratings
🗓️ 20 November 2025
⏱️ 65 minutes
🔗️ Recording | iTunes | RSS
🧾️ Download transcript
Summary
Today’s show:
The race for more compute gets all the headlines, but Mistral AI CEO Arthur Mensch tells us the REAL danger is running out of data!
On TWiST, Jason and Alex are chatting with the head of France’s largest foundation AI company, Mistral AI, about how enterprises can get more out of their AI pilots, whether the West is lacking leadership on open-source, why Mistral AI isn’t pursuing every possible kind of AI application, and optimizing your models for performance rather than benchmarks (aka benchmaxxing). Find out what it’s like to run a huge AI company in today’s intense environment!
PLUS… why everyone made the wrong call on OpenAI vs. Google… Kraken’s fresh $20 billion valuation… Jason’s urgent message to every company thinking about M&A… and much more!
Timestamps:
(3:35) Why did everyone get it wrong on Google vs. OpenAI?
(10:51) Why Jason thinks drone delivery will make everyone safer
(9:52) SQUARESPACE: Use offer code TWIST to save 10% off your first purchase of a website or domain at https://www.Squarespace.com/TWIST
(11:26) Mistral AI CEO Arthur Mensch joins us! What’s it like to run a foundation model AI company in this intense environment?
(14:50) Why enterprises aren’t getting enough value out of AI and how Mistral AI aims to correct this
(18:45) Why Mistral AI’s goal is, ultimately, to disappear…
(20:19) LinkedIn Ads: Start converting your B2B audience into high quality leads today. Launch your first campaign and get $250 FREE when you spend at least $250. Go to http://linkedin.com/thisweekinstartups to claim your credit.
(22:49) Are Europe and the US lacking leadership on open source AI?
(24:53) Mistral AI doesn’t need to compete in every category... how big of an advantage is focus?
(27:05) The “two bottlenecks” in Machine Learning: Compute and Data
(29:01) How Mistral AI works with human experts to annotate their models, and where they source them from
(30:06) Nexos.ai: Stop Shadow AI in its tracks with the unified platform for secure AI adoption and productivity. Try it with a free 14-day trial at https://nexos.ai/twist.
(34:24) Is optimizing your model for benchmark tests (aka benchmaxxing) like cheating?
(40:08) Will we ever be able to run smaller AI models on our phones?
(44:21) When does Arthur think we’ll have truly helpful humanoid robots in our lives?
(49:42) Suno lawsuits and why Jason says you should NEVER break the rules, ESPECIALLY with the music industry
(54:28) Release the Kraken! They’re now worth $20 billion.
(56:48) Jason is going to F1… should he take up the offer and do a dry run?
(58:30) How Meta beat the FTC and why Jason is so excited about it
(1:01:37) Who should buy Crunchbase?
Transcript
Click on a timestamp to play from that location
| 0:00.0 | But year, an AI model can do drive any city, drive anywhere from in Europe, Madrid to Moscow, |
| 0:07.3 | and now it's safely 100 out of 100 times, 1,000 out of 1,000 times. |
| 0:11.5 | Pick a year. |
| 0:12.2 | Territically, I think it would actually already work. |
| 0:14.7 | 10,000 out of 10,000 times perfectly? |
| 0:16.9 | No, no, probably not. |
| 0:18.1 | Some folks have moved to hiring experts for 50 bucks an hour, asking them |
| 0:22.7 | questions and then just having proprietary streams of knowledge. We have a company, Microa1. |
| 0:26.9 | How do you think about hiring experts to build out the knowledge base? |
| 0:32.4 | What we found is that onboarding and having full-time employees that actually have the expertise to judge whether we're |
| 0:38.7 | actually making progress is super important. There's still an amount of information that you're |
| 0:43.3 | not going to get to whatever you pay for. If you're not partnering with a company who actually |
| 0:47.5 | has the knowledge, you need to source people that are experts in their fields, usually PhDs, |
| 0:53.5 | and have an interest for computer science. |
| 0:55.2 | If you find these two things, suddenly you have someone who is interested in driving the |
| 0:59.3 | competence of an AI model forward in their field. |
| 1:02.5 | This week in startups is brought to you by Nexus. |
| 1:06.0 | Stop Shadow AI in its tracks with the unified platform for secure AI adoption and productivity. |
| 1:12.4 | Try it with a free 14-day trial at nexus.a-I slash twist. |
| 1:17.4 | LinkedIn ads. Start converting your B2B audience into high-quality leads today. |
| 1:22.4 | Launch your first campaign and get $250 free when you spend at least $250.50. |
| 1:28.4 | Go to LinkedIn.com slash this week in startups to claim your credit. |
... |
Transcript will be available on the free plan in 1 days. Upgrade to see the full transcript now.
Disclaimer: The podcast and artwork embedded on this page are from Jason Calacanis, and are the property of its owner and not affiliated with or endorsed by Tapesearch.
Generated transcripts are the property of Jason Calacanis and are distributed freely under the Fair Use doctrine. Transcripts generated by Tapesearch are not guaranteed to be accurate.
Copyright © Tapesearch 2025.