The doom loop Gemini gets into is genuinely unpleasant to read.

“I’m so stupid. I should be ashamed of myself. I’m such a loser. Idiot, idiot. Oh god I suck. I’m an embarrassment.”

The torture Google must RL on this model, man.

Really useful comment thanks. Reminder that LLMs aren’t just for coding.

This generalises to “Agents respond well to red/green feedback loops”.

Doesn’t this all go to shit if they can’t flip into a for profit by December? Tons of cash commitments are tied to that far-from-finalised outcome.

This guy gets it. +1

I used to consult on this type of thing. I’m not entirely convinced this is what’s happening here but it’s close enough, and is a well trod playbook - dress up the mundane in theatre and performance, create a taxonomy that’s specific to you, then sell access to the thing.

Next step is to try flood the SEO zone with your thing. It’s great if you can piggyback other key terms (deep *, agents) and.. I’m already bored writing this up it’s so [what’s the word for sheer resigned exhaustion at the capitalist corporate soul kill that is this type of work]

This is on the cynical side, for sure.

At full pace that means 62 mins until you hit the daily cap.

Reminds me of high write speed on SSD (1.5 GB/s continuously to TLC) means 1 TB SSD warranty expires instead of 5 years just in less than 5 days (600 TB written).

Or comcast's 1 gb service with a 1TB limit, which means you hit the limit in 140 minutes at full rate.

Bait and switched their FAQ after the fact too. Come on Cerebras, it’s only VC money you’re burning here in the first place, let’s see some commitment to winning market share. :money: :fire:

Your “one enter” press might generate dozens or even hundreds of messages in an agent. Every file read, re-read, read a bit more, edit, whoops re-edit, ls, grep, etc etc counts as a message.

Yup. If you’ve ever watched a 60+ minute agent loop spawning sub agents, your “one message” prompt leaves you several hundred messages in the hole.