
Connor
189 posts

Connor
@ConnorTalksAI
High level basketball player Learning and building with AI Future Software Engineer 17
शामिल हुए Nisan 2026
36 फ़ॉलोइंग51 फ़ॉलोवर्स

@gitfate @bridgebench that’s actually really cool - so it’s less beginner-friendly but is good when prompted correctly?
English

@ConnorTalksAI @bridgebench GPT is good at front end, and can even pass Claude. The issue is that nobody knows how to use GPT properly. You can image generate the design and tell GPT to implement it. Something Claude cannot do.
English


@explorersofai I don’t think so - I feel like Anthropic would say it’s AGI if it’s even close because it would be good for marketing, right?
English

@trenchesbase @BatsouElef There are lots of tools for that like ElevenLabs, hope that helps!
English

@marksikaundii @BatsouElef I think it’s really going to improve with Grok 4.5 or Grok 5, but it’s lower safeguards are a reason that it is willing to give more unhinged content
English

@BatsouElef How good is grok for coding honestly I have never used it, I know grok is good when it comes to roast others😂
English

@Skymography This is crazy! Out of curiosity, do you do most of your animations with Claude tools or by hand?
English

Recent work for Anthropic, was the lead animator on this 🕺🏻
Anthropic@AnthropicAI
New Anthropic research: A global workspace in language models. Of everything happening in your brain right now, only a tiny fraction is consciously accessible—thoughts you can describe, hold in mind, and reason with. We found a strikingly similar divide inside Claude.
English

@yashhq_22 if you have the idea and do the marketing - then definitely yes
English

I keep thinking about the idea of “brain scanning” an AI model.
Not just looking at the prompt and output, but recording the activations inside the network while it thinks.
Then:
Start with one neuron.
See what it responds to.
Change its value and watch the output shift.
Then move up to groups of neurons, layers, and eventually whole activation patterns.
Could we train another model to translate those patterns into human-readable concepts?
Basically: instead of asking an AI what it thinks, study the internal signals that caused the answer.
Maybe the future of AI interpretability looks less like reading outputs and more like mapping an artificial brain.
English

@AnthropicAI Could the idea to look inside of LLM thought processes relate to backpropogation? You could do a backward pass after each forward pass and then adjust neural network weights to find what neurons affect what words
English

@Itsbrutox I don’t understand why it’s so hard to recode an iOS app into an Android app. I understand that they use two different languages and that is requires effort but I don’t understand why it’s this much
English

@Heyyanyaa what is your strategy? I have lots of consistency but I don’t know how to have strategy
English

@themishra4402 I’m going to be honest - I still am not 100% sure what git is, what does it mean?
English

@Surendar__05 I think that’s because Apple understands that gaming will be dominated by Windows (more games support it) and also that high level gamers like to build PC’s.
With that being said, Apple devices are certainly good enough to run most games that are compatible with them
English

@astropol0 To be honest, the next Grok model could be special - there is one training with I think around 5-7 Trillion Parameters
English

depends whether I actually have customers yet and whether those customers pay money - I want to please paying customers as that is how you grow a business but starting out an app you should use personal preference and friend recommendations (saying this like I’ve ever actually shipped an app and got users 😭😭)
English






















