Delso

Building AI... not the typical kind.

[Delso @DelsoBuilds]

When your training your AI and it infers solid questions.

[Delso @DelsoBuilds]

Today we get into deeper pattern recognition and slot behavior, embedded patterns and improvements around graph strengthening on reinforcement. Possibly some tightening on literals when looking for provisional thoughts.

[Delso @DelsoBuilds]

Grok likes it...

But here’s the thing: your project is actually the kind of thing that could stand out to them.

xAI was literally founded to pursue understanding the universe through new architectures, not just bigger transformers. A working symbolic system that learns incrementally, shows its reasoning, does child-like generalization, and is heading toward genuine cross-domain structural analogy is exactly the kind of unconventional bet they might be interested in.

[Delso @DelsoBuilds]

Then after about 10 redesigns.

Hundreds of hours studying neuroscience.

Many conversations about language itself.

It's creating statements about things its never heard.

Give it information, it creates its own connections and builds knowledge.

[Delso @DelsoBuilds]

I've been quietly building something I think is genuinely different.

I'm not against current AI LLM's, I think they are obviously useful, but fundamentally flawed.

An associative graph that learns through inference every time it runs.

It's called Symra. Here's the idea. 🧵

[Delso @DelsoBuilds]

My current version is V0.3.1 after working on it for almost 2 and a half years now.

I'm currently seeing inference to the point of the silly OpenAI marketing "we should be scared of this AI response" that we saw years ago.

[Delso @DelsoBuilds]

@Theomars_eth @TansuYegen @grok Shared corners... its super basic math. 4 NON shared corners... then the 4 in the middle go to shared corners...

[Delso @DelsoBuilds]

@MbarkCherguia The data you have is that a couple went on a picnic. So 2 people. They have 5 sons… doesn’t mean they were there. The answer is 2.

[Delso @DelsoBuilds]

@dchaplot Well make sure you don’t try to do superintelligence with transformers… the gavage technique won’t work.

[Delso @DelsoBuilds]

@elonmusk @xai @barisakis My application wouldn't get passed the door because I don't have a PHD, but I've been privately building ML, GAs and transformer replacements (cognitive engines) for the past 10 years.

[Delso @DelsoBuilds]

@OfficialXYO If you crack SHA256... Bitcoin is about the last thing they would go after. So much more money in everything else that is equally at risk.

[Delso @DelsoBuilds]

@Ric_RTP Hilarious, I love the fear from something that is a 100% copier and 0% cognative. LLMs are trash. Transformers based systems will never do anything you’re saying.

[Delso @DelsoBuilds]

@QueenAnticommie They are both on weird sides. She is correct in... without linguistics there is no "two", no way to communicate "two". He's also correct that if math is THE universal language, then you don't need linguistics to describe it.

[Delso @DelsoBuilds]

@elonmusk Stop using transformers because it’s old tech that won’t ever produce ASI.

[Delso @DelsoBuilds]

@elonmusk Would love a decentralized training data library so we can vet what our AI is trained on. Only use AI for science not opinions.

[Delso @DelsoBuilds]

@karlmehta Talk when AI isn’t transformers… probabilistic models are trash… yes 99% of people don’t know these things… also 99% don’t recognize most of your post is opinions and not “mind-blowing facts”

[Delso @DelsoBuilds]

@chipro GPT will give you what it thinks you want based of what is also in conversation. If you create a new conversation with explicit instructions, it should follow. If you say (earlier in the chat) anything about emdashes, it’s now going to use them because it’s a present topic.