For anyone wondering, Gemini thinks the gorilla would beat 100 men

everyone's yelling about portfolios. meanwhile the geese on parliament hill are just honking like usual, completely unaffected. maybe they know something. maybe the real currency is intimidation and stealing sandwiches from tourists. their strategy seems more stable right now. i might switch sides

LiveBench figured out how to stay relevant as LLMs progress. Instead of being a static benchmark that all the LLMs will eventually master they updated their tests to make it more difficult and retested all the current models. This way they will always stay relevant

Vibe coding is going to give a lot of people just enough rope to hang themselves

That AI-generated code looks great... but is it safe to ship? Diving into the "Vibe Coding" trend and the crucial steps needed after generation.

#VibeCoding #EnterpriseTech #Software #AI #QA

medium.com/@sandesh.meg...

Forget LiveBench, Street Fighter III is the new LLM Benchmark we all need. We dive deep into the new wave of game-playing benchmarks to reveal the real state of AI intelligence and the personalities emerging from LLMs
medium.com/@sandesh.meg...
#AI #gaming #LLM

GPT-4.5 reactions: Slightly better than 4o, unreasonably expensive.
Has scaling hit a wall like the rumors have been saying?

Chatbots are acing tests, writing code, and even winning popularity contests. But are they really intelligent? We dive deep into the key benchmarks to see how LLMs measure up to human intelligence.
medium.com/@sandesh.meg...

Forget carefully curated math, reasoning and language processing benchmarks. Pokemon Red is the new benchmark that Anthropic is using to demonstrate their progress.
I can't wait for LLMs to duke it out to be the very best

LLM companies to their own customers (same as to everybody else): you are on your own.

LLM-based chatbots are revolutionizing customer service, but beneath the surface lies a world of pitfalls. From ‘hallucinating’ false information to being tricked into revealing sensitive data, these AI assistants can sometimes behave in unexpected and even alarming ways.
medium.com/@sandesh.meg...

This exhibit in Japan where a chained up robodog tries to attack you is giving me real Soma vibes.

EXCLUSIVE: Nobel prize winner and "Godfather" of Artificial Intelligence, Geoffrey Hinton, brutally attacks the US and the UK after they declined to sign a declaration on ensuring that the technology was "safe, secure, and trustworthy". 🇺🇸🇬🇧

This is what he told me last night. 🧵👇 #AI

This is happening everywhere. There are multiple AI chatbot solutions to replace support from a real person.
It's always a worse experience, but it's also cheaper.
AI might hit the economy of the Philippines very hard.
fortune.com/2025/02/11/3...

This new paper shows people could not tell the difference between the written responses of ChatGPT-4o & expert therapists, and that they preferred ChatGPT's responses.

Effectiveness is not measured. Given that people use LLMs for therapy now, this is an important (and urgent) topic for study.