Mastodawn

poleguy looking for lost tools Apr 20

I just got access to Claude Opus 4.7... it burns tokens at a 7.5x rate...

Our recent department meeting told us to use the AI tools to "be efficient," or else no added head-count.

They aren't actually measuring efficiency. But they are looking at token usage.

I think they are presuming token usage is going to be correlated with efficiency and want "number go up" metrics.

So my conclusion is that I should immediately switch to Claude Opus 4.7 to burn tokens faster.

#AIslop #AIcritic

Show thread

poleguy looking for lost tools Apr 17

... I'm not sure I'm happy with #LLM's burning the ocean, with the "AI" #software running as a service and funneling all the money to the top. I'm not sure I want my "personal" computer being effectively owned by someone else.

But wow, those RSI inducing editing tasks I struggled with in the 90's are just the tip of an entire class of essentially solved problems.

The machines can almost operate as I had foreseen and hoped. And yet... and yet... I have no more words.

#aislop #AIcritic

Elena Rossini on GoToSocial ⁂Apr 1

just leaving this here...

🔗: https://bsky.app/profile/jay.bsky.team/post/3micqcyeawc2g

#Bluesky #SameOldSameOld #OpenWashing #NoAI #AIcritic

poleguy looking for lost tools Mar 21

I just read this article about chess-like vs poker-like problems, and how poker-like problems are bad candidates for solution with LLM type models.

I'm looking for more reading like this about the underlying architecture, behavior, and future directions of the models and the cutting edge research around them. Does anyone have links they found valuable?

https://www.latent.space/p/adversarial-reasoning

#noai #aicritic #ai #llm

Experts Have World Models. LLMs Have Word Models.

Most expert work isn’t “produce a probable artifact”; it's "choose a good move considering other agents, guessing hidden state". LLMs default to single-shot artifacts and need World Models to progress

Latent.Space

Elena Rossini ⁂Mar 17

Yesterday my VPS set off a warning, as it was hit by a huge spike in incoming traffic, peaking at 55GB at 2:15pm and lasting for an hour.

Upon investigating, it turns out it was my PeerTube instance that was targeted.

Where did the traffic come from?

meta-externalagent (aka Meta's web crawler which is used to grab content to train its AI system).

I feel a little bit violated thinking my Fediverse promo video was grabbed by it, sigh.

#AIcritic #NoAI

poleguy looking for lost tools Mar 13

New blog post.

https://poleguy.com/blog/john-henry-and-the-steam-hammer.html

#AI #NoAI #AIcritic #slop #FPGA #verilog #vhdl

Show thread

bsmall2 Jan 19

🧵
> ... how to be a good AI critic.. “How to be a critic whose criticism inflicts maximum damage on the parts of AI that are doing the most harm.”
> .. AI can’t do your job. It can help you do your job, but that does not mean it is going to save anyone money.
> Take radiology: there is some evidence that AI can sometimes identify solid-mass tumors that some radiologists miss... no one is investing hundreds of billions in AI companies because they think AI will make radiology more expensive, not even if that also makes radiology more accurate.
#AICritic