@thecaptain_nemo
It makes sense if you consider that most sci-fi is unscientific.
Like Star Wars, which is really just fantasy in space. It doesn't even try to create believable tech. It's just magic.
So I can spare you watching the pod with
@Sama
.
It's boring AF.
Absolutely nothing new is revealed. 0.
Just blah blah, love love.
Watch the pod with LeCun instead, it's great.
We’ve developed Rule-Based Rewards (RBRs) to align AI behavior safely without needing extensive human data collection, making our systems safer and more reliable for everyday use.
People claiming that GPT-5 can only be released "after the election" :
Realistically a freaking LLM has no impact on elections at all. It's negligible.
Google has 1000x more impact by messing with the search results. That's a real danger and should be taken seriously.
@kimmonismus
The greatest safety blog of all times.
The blog to redefine blogging as we know it.
Nothing will be as it used to be. A new blogging age starts.
Big day today .
It's Thursday and we might get another safety blog from OpenAI.
"About the dangers of eating unripe strawberries", is the rumored title.
so
@iruletheworldmo
was wrong... strawberry did not get announced or released last night... and yet instead of you all blocking him he is now at almost 10k followers overnight from 3k... what are yall doing bro...
Little birdies told me 🍓 / Q* hasn’t released yet as they aren’t happy with the latency and other little things they want to further optimise.
Not sure if big patience or small patience.
Either way,
@ChatGPTapp
Shhh patience 🤫 🤫 🤫
Prediction:
Techniques like Q* sitting on top of LLMs (for improved reasoning) will be challenging for open source to reproduce.
Having the weights of a model will only get you so far. Now you also need to replicate the right inference scheme which can be quite complex.
I want to bring attention to this pastebin by an anon, before he will delete his account next week.
In the text it's being claimed that OpenAI is developing a new kind of AI model, trained on *computer games*, with an RL approach.
Thoughts?
To be fair, this explains to me a lot why GPT-4o-advanced voice is delayed. Just imagine giving it to your children and then this happens.
Because some people don’t get it: after the „NO“ GPT cloned her voice by its own and speaks in her voice. What the..
Edit2: yes, it’s valid
Big day today .
It's Thursday and we might get another safety blog from OpenAI.
"About the dangers of eating unripe strawberries", is the rumored title.
ChatGPT got a silent upgrade to a new model !
It's not the same model as gpt-4o-2024-08-06
but It's another new model with several upgrades
Very mysterious 🤔
Reminder that e/acc is just a very loosely organized troupe of mostly anons.
There's no guarantee for authenticity of any communication coming from "members", except for maybe a couple of key people.
Always good to keep a healthy dose of skepticism.
ChatGPT got a silent upgrade to a new model !
It's not the same model as gpt-4o-2024-08-06
but It's another new model with several upgrades
Very mysterious 🤔
Tesla FSD is going to use an End-To-End neural network architecture, just like
@comma_ai
is using. Interestingly this move was predicted by
@realGeorgeHotz
long ago on the
@lexfridman
podcast.
The infamous GPT-5 countdown account posted an apology / explanation for the nothingburger on June 6.
(Personally I think it's a good gesture, better than just deleting the account)
Predictions for topics in upcoming
@Sama
at Lex pod:
* AI Safety
* Elections and AI Safety
* Risks of AI
* Elon and OpenAI
* LOVE. LOVE. LOVE.
That's it! Hope you didn't expect some new release.
👍
AI Safety and AI Alignment have the opposite effect and can make things non-safe
According to Richard Sutton, one of the founders of reinforcement learning
An actual AI expert.
Richard Sutton says AI safety advocates are creating the opposite of what they seek - we should not try to solve the AI control problem because a decentralized system of independent agents is better than everything being aligned under one goal
Jensen showing us the endgame of GenAI:
Multi-modal in -> Multi-modal out.
The is the final model we will all use one day.
Can get any modality and generate every modality.
While also being able to work without each of it's parts.
We’re testing SearchGPT, a temporary prototype of new AI search features that give you fast and timely answers with clear and relevant sources.
We’re launching with a small group of users for feedback and plan to integrate the experience into ChatGPT.
It's incredible how many features and products OpenAI has in the pipeline
...but somehow those take forever to roll out.
Like the ChatGPT UI hasn't seen any improvements in a long time.
Curiously, while there were visualizations comparing Anthropic & OpenAI in terms of velocity of newly added features, there are many new features in the works for ChatGPT not launched yet (at least those that I know about) - not sure if these are not ready for prime-time yet or
A truly end-to-end image AI model should unlock amazing image editing capabilities!
"Take this photo here, remove the tree, make me smile, and render the colors more brilliant."
Unfortunately we won't get this for "safety" reasons 🙄
Hot take, I want the promised 4o image capabilities more than voice.
Something magic about the prompt accuracy and working in tandem with an alien ai to create.
“Art is the highest form of hope. What we are doing here is hope work.“