Single brain implant restores bilingual communication to paralyzed man
https://arstechnica.com/science/2024/05/single-brain-implant-gives-paralyzed-man-bilingual-communication/ #psychology #BrainImplant #biliungual #communication #BRAVO #GPT2

Before launching, GPT-4o broke records on chatbot leaderboard under a secret name - Enlarge (credit: Getty Images)

On Monday, OpenAI employee Will... - https://arstechnica.com/?p=2024084 #largelanguagemodels #multimodalmodels #machinelearning #simonwillison #chatbotarena #gpt2-chatbot #gpt-4-turbo #aivibes #chatgpt #chatgtp #biz⁢ #gpt-4o #openai #gpt-4 #lmsys #ai

Mysterious “gpt2-chatbot” AI model appears suddenly, confuses experts - Enlarge (credit: Getty Images)

On Sunday, word began to spread... - https://arstechnica.com/?p=2020588 #machinelearning #simonwillison #aibenchmarks #chatbotarena #ethanmollick #gpt2-chatbot #samaltman #aivibes #gpt-3.5 #gpt-4.5 #biz⁢ #openai #gpt-3 #gpt-4 #gpt-5 #lmsys #ai

There is a mysterious new chatbot from OpenAI on https://chat.lmsys.org/ , it's called GPT2 not to be confused with the old model with the same name.
This models seems to do several things better then GPT-4.
Everybody is speculating what it is and what it is not.
#GPT2 #OpenAI #AI

@worldai #llmsys #gpt2 #gpt5 #genai

Introducing GPT-5?

Mysterious GPT2-Chatbot Outperforms GPT-4!

https://youtu.be/u16ipSeYH7U?feature=shared

(Ed : Who did this #OpenAi #MSFT #Apple feels like some used a higher model to train a GPT2 )

How to break an AI (the illustrated guide )

I am posting this for fun, to show how fragile #AI systems are, and how ridiculous it is to imply that they are intelligent or could wipe us out.

1) Grab a model. For this demo, I will take GPT2 because it fits on my laptop.

2) Copy-paste code for running and fine-tuning the AI. You can take mine here, which will also download #GPT2 for you: https://github.com/possible-worlds-research/AI-buster.

Optional: see how the model, for now, is working as it should… 1/4

Glitch Tokens

https://www.youtube.com/watch?v=WO2X3oZEJOA

Brain vs #LLMs? Hierarchical predictions integrating nonlinguistic & linguistic knowledge provide a more comprehensive account of behavioral & neurophys response to #speech, compared to next-word predictions by #GPT2 @YaqingSu @labgiraud #AI #PLOSBiology https://plos.io/409yGSS

Brain vs #LLMs? Hierarchical predictions integrating nonlinguistic & linguistic knowledge provide a more comprehensive account of behavioral & neurophys response to #speech, compared to next-word predictions by #GPT2 @YaqingSu @labgiraud #AI #PLOSBiology https://plos.io/409yGSS

Brain vs #LLMs? Hierarchical predictions integrating nonlinguistic & linguistic knowledge provide a more comprehensive account of behavioral & neurophys response to #speech, compared to next-word predictions by #GPT2 @YaqingSu @labgiraud #AI #PLOSBiology https://plos.io/409yGSS

If you think #ChatGPT based on #GPT3 is so powerful, think about its predecessor, #GPT2, which was "too dangerous to release" back in 2019... https://www.independent.co.uk/tech/ai-artificial-intelligence-dangerous-text-gpt2-elon-musk-a9192121.html

Fun fact: the article's URL contains `elon-musk`. No mention of Musk in the article whatsoever...

Generate Mario levels with GPT2 trained model https://github.com/shyamsn97/mario-gpt You could use generative AI to create levels for your game! Of course you will have to play them and tweak them... #generativeai #ai #mario #gpt2

Happy to share our new paper “Language model acceptability judgements are not always robust to context” https://arxiv.org/abs/2212.08979! We prepend several kinds of context to minimal linguistic #acceptability test pairs and find #LMs (#OPT, #GPT2) can still achieve strong performance on BLiMP & SyntaxGym, except in some interesting cases. [1/7]

Interesting in case you suspect that some student work is #AI-generated: The #GPT2 Output Detector Demo shows to what extent a text was generated by #ChatGPT.

https://huggingface.co/openai-detector/

#ICYMI

In case you want to experiment with (so-called) #AI models, a lot of them have been released by https://huggingface.co/

all available directly via #python packages

for example #gpt2 for machine-generated absurdity (#dada) https://huggingface.co/gpt2

No #GPU required, works just fine on a normal CPU

I was able to get #OpenAI’s #GPT2 running locally yesterday. Getting coherent results from #ChatGPT and #GPT3 is amazing, but half the fun with AI writing back in the day was how high the writing-style was.
GPT3 just gets boring after a while, especially when you consider the fact you can’t run it locally and some content gets completely censored. Something feels amazing about being able to just do it on your own laptop without any cloud involved.

There's a GPT-2 Output detector from https://huggingface.co/ - it shows you how likely it is, that a text has been generated with ChatGPT or OpenAI
#gpt #gpt2 #openai
https://huggingface.co/openai-detector/