aichannel

sonnet reasoning would be the strongest model, no? 

is reasoning more than stacking a regular llm with chain of thought?

product @seedclub. i host S-TIER @ X Spaces & i interview crypto builders @ web3galaxybrain.com.

Sonnet still currently the best for most stuff. Cost v performance R1 is fine though dependent on use case. 

And yes all these LLMs and “agents” are just stacks of COT prompts. 

You should read about COCONUT though. It may give R1 a run for its money.

oh god no sonnet has fallen off heavily. terrible.

Haven’t seen that degradation personally but rarely use it directly tbf. 

Multi-model inference pipelines are like the only way to get solid output

i literally all caps at sonnet in cursor now because it makes mistakes so ridiculous

prev founded rocketr.net, swish.id now exploring aiXcrypto identity ai. expired sprint car driver & energy data engineer. idol worship on shoni_eth.X

Ah yeah that’s nothing new but it wouldn’t surprise me if it got worse lol. 

I stopped relying on it for most coding tasks a while back. It can help accelerate repetitive tasks or frontend work but I don’t trust it for anything at the server or system level. It quite literally can’t even do basic decimal calculations (a limitation of COT reasoning). 

It’s not taking anyone’s job (yet).

for the past few weeks on my end at least... i also just hate that anthropic is such an over censored company/product tbf