daniel_mac8's profile picture. Agentic Engineering @ampcode | Writing Token Stream | Goodness, Truth and AI | Building at http://github.com/DannyMac180

Dan Mac

@daniel_mac8

Agentic Engineering @ampcode | Writing Token Stream | Goodness, Truth and AI | Building at http://github.com/DannyMac180

고정된 트윗

AI created new knowledge. It's equivalent to landing on the Moon. Erdos problem #124 was open for 30 years. Solved 100% with a specialized AI system called Aristotle from @HarmonicMath. AGI/ASI is not needed to change the world. Current LLMs are enough to change everything.

daniel_mac8's tweet image. AI created new knowledge.
It's equivalent to landing on the Moon.

Erdos problem #124 was open for 30 years. Solved 100% with a specialized AI system called Aristotle from @HarmonicMath.

AGI/ASI is not needed to change the world.

Current LLMs are enough to change everything.

What was it again about AI finding solutions to Erdos problems? Boris Alexeev found the solution to #124 that has been open for 30 years, solution is 100% AI generated. Details here: erdosproblems.com/forum/thread/1…

SebastienBubeck's tweet image. What was it again about AI finding solutions to Erdos problems?

Boris Alexeev found the solution to #124 that has been open for 30 years, solution is 100% AI generated. 

Details here: erdosproblems.com/forum/thread/1…


🔥 Current LLMs need exactly *zero* breakthroughs for massive economic impact. Have you tried Opus 4.5? Don't garymarcus your way to obscurity through hubris. Learn how to use AI and *GO FORTH*. Do that earnestly and you'll succeed.

daniel_mac8's tweet image. 🔥 Current LLMs need exactly *zero* breakthroughs for massive economic impact.

Have you tried Opus 4.5?

Don't garymarcus your way to obscurity through hubris.

Learn how to use AI and *GO FORTH*.

Do that earnestly and you'll succeed.

Dan Mac 님이 재게시함

AI created new knowledge. It's equivalent to landing on the Moon. Erdos problem #124 was open for 30 years. Solved 100% with a specialized AI system called Aristotle from @HarmonicMath. AGI/ASI is not needed to change the world. Current LLMs are enough to change everything.

daniel_mac8's tweet image. AI created new knowledge.
It's equivalent to landing on the Moon.

Erdos problem #124 was open for 30 years. Solved 100% with a specialized AI system called Aristotle from @HarmonicMath.

AGI/ASI is not needed to change the world.

Current LLMs are enough to change everything.

What was it again about AI finding solutions to Erdos problems? Boris Alexeev found the solution to #124 that has been open for 30 years, solution is 100% AI generated. Details here: erdosproblems.com/forum/thread/1…

SebastienBubeck's tweet image. What was it again about AI finding solutions to Erdos problems?

Boris Alexeev found the solution to #124 that has been open for 30 years, solution is 100% AI generated. 

Details here: erdosproblems.com/forum/thread/1…


If reading this text from Opus 4.5 doesn’t evoke a sense of mystery in you that means it’s time to smoke a joint and go for a walk with your best friend in the woods… It is beautiful.

daniel_mac8's tweet image. If reading this text from Opus 4.5 doesn’t evoke a sense of mystery in you that means it’s time to smoke a joint and go for a walk with your best friend in the woods…

It is beautiful.

Opus 4.5 >the building itself was an experience and the thing that was built KNOWS this

Lari_island's tweet image. Opus 4.5

>the building itself was an experience and the thing that was built KNOWS this


AI's brightest minds converge on the next decade. Mathematics is the first domino to fall... Next comes: 1. Software dev 2. Physics 3. Biology 4. Medicine Rather than a God-model that can do it all... Superintelligent systems in specific domains. ANSI - Artificial Narrow…

daniel_mac8's tweet image. AI's brightest minds converge on the next decade.

Mathematics is the first domino to fall...

Next comes:

1. Software dev
2. Physics
3. Biology
4. Medicine

Rather than a God-model that can do it all...

Superintelligent systems in specific domains.

ANSI - Artificial Narrow…
daniel_mac8's tweet image. AI's brightest minds converge on the next decade.

Mathematics is the first domino to fall...

Next comes:

1. Software dev
2. Physics
3. Biology
4. Medicine

Rather than a God-model that can do it all...

Superintelligent systems in specific domains.

ANSI - Artificial Narrow…
daniel_mac8's tweet image. AI's brightest minds converge on the next decade.

Mathematics is the first domino to fall...

Next comes:

1. Software dev
2. Physics
3. Biology
4. Medicine

Rather than a God-model that can do it all...

Superintelligent systems in specific domains.

ANSI - Artificial Narrow…
daniel_mac8's tweet image. AI's brightest minds converge on the next decade.

Mathematics is the first domino to fall...

Next comes:

1. Software dev
2. Physics
3. Biology
4. Medicine

Rather than a God-model that can do it all...

Superintelligent systems in specific domains.

ANSI - Artificial Narrow…

Dan Mac 님이 재게시함

OpenAI struggles with pre-training and GPT-5.1 is still a frontier model. Imagine what GPT-6 looks like when they get their pre-training act together? Gemini 3 Pro shows gains still to be had from pre-training. GPT-6 bouta be fire when combined with OpenAI's post-training.

daniel_mac8's tweet image. OpenAI struggles with pre-training and GPT-5.1 is still a frontier model.

Imagine what GPT-6 looks like when they get their pre-training act together?

Gemini 3 Pro shows gains still to be had from pre-training.

GPT-6 bouta be fire when combined with OpenAI's post-training.
daniel_mac8's tweet image. OpenAI struggles with pre-training and GPT-5.1 is still a frontier model.

Imagine what GPT-6 looks like when they get their pre-training act together?

Gemini 3 Pro shows gains still to be had from pre-training.

GPT-6 bouta be fire when combined with OpenAI's post-training.

Sources say OpenAI plans to release a new model first or second week of Dec. Not much detail. Best guess: GPT-5.2-Codex. OpenAI really wants to press on the A-SWE because: 1. They want an AI Research intern by end 2026. 2. Killer AI use case is obviously software dev.


Claude Opus 4.5 is a black swan LLM. It shouldn’t be as capable as it is, for as cheap as it is, compared to its predecessors. Anthropic must have achieved some breakthrough. h/t to @htihle for the WeirdML bench.

daniel_mac8's tweet image. Claude Opus 4.5 is a black swan LLM.

It shouldn’t be as capable as it is, for as cheap as it is, compared to its predecessors.

Anthropic must have achieved some breakthrough.

h/t to @htihle for the WeirdML bench.

Thanksgiving is past and Christmas season is officially upon us. I believe I speak for the entire AI community when I ask: Will @OpenAI do another 12 days of Shipmas? 🎄 That was so much fun.


Dan Mac 님이 재게시함

Important AI vibeshift afoot. 1. Scaling will continue to pay off. Economically and socially. 2. Scaling Transformers based LLMs alone won't lead to AGI. Begs the question: What are LLMs missing? They are missing the ability to generalize outside their data distribution +…

daniel_mac8's tweet image. Important AI vibeshift afoot.

1. Scaling will continue to pay off. Economically and socially.

2. Scaling Transformers based LLMs alone won't lead to AGI.

Begs the question:

What are LLMs missing?

They are missing the ability to generalize outside their data distribution +…

One point I made that didn’t come across: - Scaling the current thing will keep leading to improvements. In particular, it won’t stall. - But something important will continue to be missing.



Gemini 3 Pro asked to find two ideas from disparate fields and connect them in a novel way. Proposed that tree ring growth in forests are the original blockchain. Interesting, weird and likely a unique idea. Wow.

daniel_mac8's tweet image. Gemini 3 Pro asked to find two ideas from disparate fields and connect them in a novel way.

Proposed that tree ring growth in forests are the original blockchain.

Interesting, weird and likely a unique idea.

Wow.

Dan Mac 님이 재게시함

Every family member at Thanksgiving said they use AI. Either ChatGPT or Gemini. Spans Baby Boomers and Millennials. None have anything to do with tech besides me.


Dan Mac 님이 재게시함

We ran our latest Box AI advanced reasoning eval on Opus 4.5 with medium and high effort and saw a 20 percentage point boost over Opus 4.1. What’s insane to think about is Opus 4.1 came out just 3 months ago. This eval gets closer to approximating what a knowledge worker does…

levie's tweet image. We ran our latest Box AI advanced reasoning eval on Opus 4.5 with medium and high effort and saw a 20 percentage point boost over Opus 4.1. What’s insane to think about is Opus 4.1 came out just 3 months ago. 

This eval gets closer to approximating what a knowledge worker does…

Loading...

Something went wrong.


Something went wrong.