Mistral's New Plan for Improving Its AI Models: Training Data from Enterprises (wsj.com) 11
Paris-based AI giant Mistral "is pushing to improve its models," reports the Wall Street Journal, "by looking inside legacy enterprises that hold some of the world's last untapped data reserves...."
Mistral's approach will be to form partnerships with enterprises to further train existing models on their own proprietary data, a phenomenon known as post-training... [At Dutch chip-equipment company ASML], Mistral embeds its own solutions architects, applied AI engineers and applied scientists into the enterprise to work on improving models with the company's data. [While Mistral sells some models under a commercial license], this co-creation strategy allows Mistral to make money off the services side of its business and afford to give away its open source AI free of charge, while improving model performance for the customer with more industry context...
This kind of hand-holding approach is necessary for most companies to tackle AI successfully, said Arthur Mensch [co-founder and chief executive of Mistral]. "The very high-tech companies [and] a couple of banks are able to do it on their own. But when it comes to getting some [return on investment] from use cases, in general, they fail," he said. Mensch attributes that in part to a mismatch between expectations and reality. "The curse of AI is that it looks like magic. So you can very quickly make something that looks amazing to your boss," but it doesn't scale or work more broadly, he said. In other cases, enterprises simply might not know what to focus on. For example, it is a mistake to think equipping all employees with a chatbot will create meaningful gains on the bottom line, he said. Mensch said to fully take advantage of AI, companies will have to rethink organizational structures. With information flowing more easily, they could require fewer middle managers, for example.
There is a lot of work yet to do, Mensch said, but in a large sense, the future of AI development now lies inside the enterprise itself. "This is a pattern that we've seen with many of our customers: At some point, the capabilities of the frontier model can only be increased if we partner," he said.
This kind of hand-holding approach is necessary for most companies to tackle AI successfully, said Arthur Mensch [co-founder and chief executive of Mistral]. "The very high-tech companies [and] a couple of banks are able to do it on their own. But when it comes to getting some [return on investment] from use cases, in general, they fail," he said. Mensch attributes that in part to a mismatch between expectations and reality. "The curse of AI is that it looks like magic. So you can very quickly make something that looks amazing to your boss," but it doesn't scale or work more broadly, he said. In other cases, enterprises simply might not know what to focus on. For example, it is a mistake to think equipping all employees with a chatbot will create meaningful gains on the bottom line, he said. Mensch said to fully take advantage of AI, companies will have to rethink organizational structures. With information flowing more easily, they could require fewer middle managers, for example.
There is a lot of work yet to do, Mensch said, but in a large sense, the future of AI development now lies inside the enterprise itself. "This is a pattern that we've seen with many of our customers: At some point, the capabilities of the frontier model can only be increased if we partner," he said.
Re: Penis Queef (Score:2)
Just be over already.. (Score:3, Funny)
"two paragraphs suggesting bad business ideas should stop, as sarcastically as possible."
Oh yes, by all means, let’s keep investing in terrible business ideas—because who doesn’t love the thrill of watching money vanish faster than dignity at a karaoke bar? Let’s keep pushing forward with “Uber, but for pigeons,” or “AI-powered fax machines,” because what the world really needs is yet another subscription box that sends people socks they didn’t ask for. Forget innovation—mediocrity is clearly the gold mine we’ve all been ignoring.
And please, let’s never stop brainstorming more brilliant schemes like “gluten-free water” or “solar-powered flashlights” (for nighttime use, naturally). The market desperately craves apps that remind you to breathe, or perhaps a service where you pay extra for the privilege of picking your own groceries. Honestly, bad business ideas shouldn’t stop—they should be cherished, nurtured, and written about in MBA textbooks under the chapter titled “How to Fail Faster Than Light.”
Re: (Score:2)
Watch your IP (Score:4, Insightful)
"partnerships with enterprises" (Score:3)
Nope (Score:2)
How to improve AI (Score:2)
Shut down all AI-focused data centers, close down all AI companies, confiscate the wealth of the tech bro millionaires and distribute it amongst their employees who are no longer employeed, and bask in a new golden age as we no longer have to hear about AI or have it choke our life with crap, and as suddenly we have a huge energy surplus and a grid that can handle everything that we need on it.
Well, OK, I guess for a true golden age we need to have not a petty narcissist as president and an entire political
Re: (Score:2)
The Trumpistan and China still beat them (Score:2)
The "AI" outfits over there are getting data straight from their gubbermints, for free.
Won't be able to win against that for a while.