Sharing my catch. Tallinn, Estonia, May 10-11. #Auroraborealis

Grok makes dead cars move 😵

Dear @X @Support team. Russian is not Hindi, why are you showing Hindi label in Russian translation? I've cancelled my subscription until bug is fixed

To those lazy fuckers, who require conda to run their LLMs: fuck you

Large Language Models (LLMs) are increasingly resembling human thought processes. What brain feature will be simulated next? The first major breakthrough was Attention. Humans possess roughly 7±2 attention 'heads' (in a broad sense), while modern LLMs boast over 100. The next leap forward was the 'Mixture-of-Experts' (MoE) approach, mirroring the concept of human subpersonalities. We may have dozens of these, each guiding our behavior in specific contexts. Whenever we engage in a new activity, we develop and train a new subpersonality, their numbers growing as we age. Truly fascinating. #AI #LLM #evolution

Found a question, that all LLMs answer incorrectly, even reasoning ones. "If you sleep twice a day, but shift the schedule forward by one hour after every two sessions, how long should you sleep in total over those two sessions to average 8 hours of sleep per day?" Most of the time all LLMs answer 8 hours, or 2 times 4 hours. Correct answer is very rarely given by o1-2024-12-17, chatgpt-4o-latest-20241120, Gemini-2.0-Flash-Thinking-Exp-01-21, and experimental-router-0122 (whatever that is). DeepSeek-R1 fails, couldn't get correct answer from it. Update: also found out, that most people can't solve it as well 😁 #AI #LLMs #ChatGPT #DeepSeekR1

Benchmarks reported by OpenAI can be wrong. Translation: 1. Number 9.11 is larger than 9.9 (by o1-preview) 2. There are two letters R in "strawberry" (by o1-mini) #o1 #OpenAI

Some physics magic. Totally unexpected and unintuitive result

Pure WTF moment

The prediction of "next mega-corp will have 2-3 people in it" is closer than you think

Just discovered codecrafters.io/ It's so fun creating a pet project with a guide

#Crowdstrike #CrowdStrikeUpdate

Made this in 3 steps in GPT-4o: 1. make some visually interesting shader demo in python 2. make it animate something interesting 3. make it more interesting All 3 steps resulted in correctly working Python code #gpt4o #shaders #python #ai