🔥The 9th Round of Easy Loan, Earn $40 Reward is in progress❗️ ⏰ Promotion Period: January 15th - Feburary 15th, 2025 👉 Register now and check more details at gate.io/campaigns/358

"Interdimensional Cable", shorts made with Veo 3 ai. By CodeSamurai on Reddit

Prompt Theory (Made with Veo 3) What if AI-generated characters refused to believe they were AI-generated?

Ok that's kinda nice. First time I've seen summoning a coding agent on github via just an @ mention

dwarkesh's most recent episode with sholto + trenton is one of the best resources for vibe-checking your takes on LLMs/AI and thinking on the near-future, since broadly everything they say here is both correct and good: youtube.com/watch?v=64lXQP…

pass@1 vs. cons@k vs. parallel compute vs. pass@k

AHHHHH. Average over N gives no advantage. It is definitely not fancy best of N 10 flips of a coin tells you more about the coin's weighting than one flip. AHHHHH

Yep LLMs still suck at video games. 0.5-1.6% completion rate at best Again, not surprising given that even visual tic tac toe or connect 4 is not yet a done deal

I hear he passed the Claudes interview round without using jailbreaks

Gemini solves an easy Portal test chamber in ~15 minutes. I try to control my frustration 01:01 looking at a closed door, hallucinates next puzzle thru the door 01:50 first dumb idea 03:12 realises key objective 03:41 thinks it's solved it; has not (relatable) 04:01 repeats same

Nice example of both the power and limits of general purpose AI agents. Operator: "go find a multiplayer game you can play live online right now and win it against a human" With just that prompt it found a multiplayer tic-tac-toe game, joined it... & lost operator.chatgpt.com/v/683aa2835ea0…

"Grandad, were you really there when computers first started to talk?" "That's right kiddo. I was right there in the first wave. Using them to streamline B2B SaaS"

People are posting LLM slop on specific programming language subreddits now And the best part? They can prompt them into bro/cuz lower case slop. But they can't escape the writing style — it's completely unavoidable.

o4-mini-high w search used a single reddit thread w 2 upvotes as the evidence for 4 different claims in a niche programming query I asked. Turns out thread is right. Is o4-mini gullible or wise? In information vacuum, I assume easily led astray by plausible but incorrect info?

What in the actual fuck:

If you're a decent writer writing for reasonably discerning/invested/smart readers, the em dash thing just isn't a problem. You can still use em dashes because your readers' subconscious AI detectors will be much more sophisticated than "omg em dash = AI!"