Technology

68349 readers

3378 users here now

This is a most excellent place for technology news and articles.

Our Rules

Follow the lemmy.world rules.
Only tech related news or articles.
Be excellent to each other!
Mod approved content bots can post up to 10 articles per day.
Threads asking for personal tech support may be deleted.
Politics threads may be removed.
No memes allowed as posts, OK to post as comments.
Only approved bots from the list below, this includes using AI responses and summaries. To ask if your bot can be added please contact a mod.
Check for duplicates before posting, duplicates may be removed
Accounts 7 days and younger will have their posts automatically removed.

Approved Bots

founded 2 years ago

MODERATORS

[email protected]

Why Anthropic’s Claude still hasn’t beaten Pokémon (arstechnica.com)

submitted 2 weeks ago by [email protected] to c/[email protected]

16 comments fedilink hide all child comments

all 17 comments

sorted by: hot top controversial new old

[–] [email protected] 89 points 2 weeks ago (3 children)

"PhD level intelligence"
Fails at basic tasks

I have never found an LLM model this relatable before.

[–] [email protected] 12 points 2 weeks ago

Damn. This is brutally relatable.

[+] [email protected] -6 points 2 weeks ago (2 children)

They're good for coding tho altho I'm not high-level experienced with that or anything. Just seems to get it right for the simple things I'm working on and the iterating over what you started with helps me a lot

[–] [email protected] 17 points 2 weeks ago

They're good for coding

https://pivot-to-ai.com/2025/03/19/ai-coding-tools-fix-bugs-by-adding-bugs/

[–] [email protected] 17 points 2 weeks ago (1 children)

It really isn't. At best it can generate short snippet solutions but it absolutely botches projects that require contextual knowledge of the program stack.

My job actively avoids because of the havoc it tends to cause not to mention legal issues.

[–] [email protected] 9 points 2 weeks ago* (last edited 2 weeks ago)

I think context is what's going to kill LLMs. They keep throwing hacks on top of it to try to make it understand context, but it never really "understands" it just makes it look like it is following the context by simulating a few pertinent cues. Every interaction is essentially a fresh slate with a few prompts hiding underneath to seed it with what looks like context, but trying to actually preserve the context of the model to the level that we would consider actual "intelligence" never mind long term planning and actual "thinking" would explode towards infinity so fast there are probably not enough resources in the universe to even do it once.

[–] [email protected] 18 points 1 week ago (1 children)

Repeat after me: LLMs do not have intelligence.

[–] [email protected] 6 points 1 week ago

They don't. But nevertheless, the progress they've made in a year is very impressive.

The question left to be seen is how it'll look in a year or two: hardly any improvement, or a beaten elite four?

[–] [email protected] 15 points 2 weeks ago* (last edited 2 weeks ago)

I wonder if it knows or can learn to save its state before big battles and trying to catch a legendary pokemon without Master Ball then reset if it doesn't succeed initially. I was a very conservative Pokemon player and I was practically "OCD" about planning ahead for that stuff

Even better it can "cheat"

[–] [email protected] 11 points 1 week ago

The fact that Twitch chat was able to beat Pokemon Red faster than AI's been able to get to Vermillion is pretty funny to me

[–] [email protected] 11 points 1 week ago

Claude frequently finds itself pointlessly revisiting completed towns, getting stuck in blind corners of the map for extended periods, or fruitlessly talking to the same unhelpful NPC over and over, to cite just a few examples of distinctly sub-human in-game performance.

Claude is just like me fr

[–] [email protected] 4 points 2 weeks ago (1 children)

If the graphics are a hurdle, choose a newer version. Why are they obsessed with Gen1?

[–] [email protected] 4 points 1 week ago

I wonder if it's because the game itself is the simplest. They keep adding new systems with each iteration—as well as better graphics—so maybe they want to start with the easiest.

[–] [email protected] 4 points 2 weeks ago (1 children)

So, it’s really just playing like any 5 year old.

[–] [email protected] 2 points 1 week ago

5 year olds are more capable than this