“Nobody cared about the model.” Eli the Computer Guy talks to The Tech Report’s Isaac Pound about the tech underpinning the AI hype and how the industry is moving away from reliance on frontier models. Instagram: https://www.instagram.com/the_tech_reprt/ Facebook: @TheTechReport
ADVERTISEMENT
LLMs are only truly useful when you already know the answers....
I'm not afraid of AI becoming smarter than us. I am afraid that we prove to be dumb enough to give control to the unintelligent, unreliable AI that we created
At this point Eli and Ed should start their own podcast
The Tech Report's intro music simultaneously stresses me out and gets me hyped.
can we just lose the data centers already?
I love Isaac's "WHAT" when Eli started talking about Anthropic not keeping log files. Honestly when I watched Eli's video about that before, my reaction, as an IT admin, was the same.
The thing is: it doesn't make sense to run these bigger and more expensive models because they don't have a hibernation mode. They are at 100% all the time. If local computing can solve small to medium size problems (most problems) big models aren't profitable
2:48 And that would be a giant problem if AI were treated like normal software: two alike requests could cost 100 times different and you cant predict it, because if you can, you dont need AI.
I listened to a radio show today; the call-in computer scientist, lecturer and “ai expert” stated that he doesn’t use AI at all in his personal life and thinks too much screen time is bad anyways. And the business owner who waxed lyrical about AI helping his business, when asked what he uses AI for, is mostly summarising emails and doing slide decks… lol this is the big selling point is it? Oh and one guy had an agent out there every day looking for the cheapest possible flight to a particular destination. Did it find any? No
The Age of Autocomplete is rife with hidden layers to goose the initial appearance of usefulness
frontier labs charge per token while efficient orchestration architectures move 90% of compute to local hardware cutting their revenue streams
The "bigger better" models are bigger and oftentimes not better or at least not in ways that are relavant or matter. LLMs are dumb as shit statistical guessing machines and that's all they ever will be. At the core of what they are this can not fundamentally ever be changed, they will always work off of statistical guessing no matter how many layers of "reasoning" there is.
We need all companies to have their c-suites liable for their criminal actions again
A.I doesn't actually think. You just use a probability machine to feed another probability machine.
They said it will find cure for cancer etc. We are getting another crappy ai note taking app, and a dubious promises of shipping more of stuff nobody needs.
Anthropic doesn't have basic logging capabilities. Folks, this is why and how regulations and SOPs get created.
Reasoning doesn't do shit. It just adds tokens to the prompt.
The "reasoning model" is just a more complex predictor than the original predictor. I get that these models use advanced math (groups and sets) and vast data sets, but they don't reason or think.
I feel like Eli could do great Patrick Warburton impressions.
I know almost nothing about tech, but even I recognize that LLM’s are basically just search engine helpers. Ai will change the world, but more in the way that MS Office changed the world, but not in the way the internet changed the world.