No, Mistral reported about 50% was training, 50% inference, but a more widely used model would have lower relative training to inference costs.
@emollick
-
AI Energy Usage: Grounded Facts for Classroom Discussion
By
–
I know that energy usage from AI prompts comes up in classroom discussion all the time. I hope this section in my latest post helps people provide a more grounded answer to the question, rather than just speculating or citing out-of-date information. https://
oneusefulthing.org/p/mass-intelli
gence
… -
Hallucinations in AI: Were 2025 Predictions Realistic?
By
–
Also who thought at the end of 2022 that all hallucinations would be solved by 2025? I haven't seen that prediction listed anywhere.
-
AI Progress Exceeds 2022 Expert Forecasts and Expectations
By
–
I don't think saying AI is ahead of where the vast majority of experts and forecasters expected in 2022 is very controversial. There is no doubt there is a jagged frontier of AI ability (heck, we coined that term in our paper!), but it is also clear that many of the areas you
-
AI Math Performance Surpasses 2022 Expert Forecasts
By
–
We can now say pretty definitively that AI progress is well ahead of expectations from a few years ago. In 2022, the Forecasting Research Institute had super forecasters & experts to predict AI progress. They gave a 2.3% & 8.6% probability of an AI Math Olympiad gold by 2025…
-
Building AI for Good While Mitigating Risks and Job Loss
By
–
And, not to sound like a broken record, that is why it is important to start dealing with the implications of steadily improving AI ability as a base case, including building for good uses (in education, healthcare) and mitigating risks like job loss, misuse, and other dangers.
-
AGI Timeline Debate: Continuous AI Advancement Beyond 2030
By
–
The realistic alternative to "AGI by 2030" is not "AI goes away" but rather continued advancement in an industry with large-scale economic value (at least by early revenue numbers) and many smart people focused on figuring out new approaches that might work if old ones fail.
-
Model Release Cycles: Dead Ends and New AI Paradigms
By
–
It is possible, especially on X, to get caught up in the excitement (or anxiety) of weekly model releases, judging each as "its so over" or "we are so back," but, in the larger scale, some possible dead ends (pre-training?) and new paradigms explored (reasoners, RL) look normal.
-
Chip Density Growth: Multiple Paths Forward Beyond Current AI Limits
By
–
60 years of exponential growth in chip density was achieved not through one breakthrough or technology, but a series of problems solved and new paradigms explored as old ones hit limits. I don't think current AI has hit a wall, but even if it does, there many paths forward now.
-
AI interfaces advancing rapidly in coding, lacking elsewhere
By
–
There are probably other workflows and UXes that might work even better. Kind of a shame that the only place where AI interfaces are advancing rapidly are in coding (not surprising given who is making AI interfaces right now).