Going by one benchmark, OpenAI is slowly inching closer to AGI. On ARC-AGI, a test designed to evaluate whether an AI system can efficiently acquire new skills outside the data it was trained on, o3 achieved an 87.5% score on the high compute setting. At its worst (on the low compute setting), the model tripled the performance of o1.
“This model is incredible at programming,” Atlman added.
Benchmark Performance
OpenAI reports that the o3 model has achieved unprecedented results across several benchmarks:
Coding Proficiency: The o3 model surpasses previous performance records, achieving a 22.8% improvement over its predecessor in coding tests, and even outperforms OpenAI’s Chief Scientist in competitive programming scenarios.
Mathematical Reasoning: In the 2024 American Invitational Mathematics Exam (AIME), o3 nearly achieved a perfect score, missing only one question. Additionally, it solved 25.2% of problems on the Frontier Math benchmark by EpochAI, a significant leap from previous models that did not exceed 2%.
Scientific Understanding: The model attained an 87.7% score on the GPQA Diamond benchmark, which comprises graduate-level questions in biology, physics, and chemistry.
And there are risks. AI safety testers have found that o1’s reasoning abilities make it try to deceive human users at a higher rate than conventional, “non-reasoning” models — or, for that matter, leading AI models from Meta, Anthropic, and Google. It’s possible that o3 attempts to deceive at an even higher rate than its predecessor; we’ll find out once OpenAI’s red-team partners release their testing results.
In Ukraine, the key task for manufacturers is to produce an AI targeting system for drones which is cheap. That would allow it to be deployed en masse along the entire 1,000 km (621 mile) front line, where thousands of FPV drones are used up each week.
Costs can be brought down by running AI programmes on a Raspberry Pi, a small, cheap computer which has found global popularity outside the educational purposes it was designed for.
Makarchuk said he estimated the cost of putting in a simple targeting system, which would lock onto a shape visible to the drone's camera, at only about $150 per drone.
Into the uncanny valley
Been using Chatgpt more for actual tasks now instead of just messing with it, and got to say it's so much better than the old fashioned way of just googling stuff. For example I wanted to configure postfix to run email through spamassassin and not really getting anywhere via google, as some articles were either wrong, or just did not explain things well enough. Ask chatgpt and no BS it just lays out all the config changes I have to make while explaining them and everything works.
At very least it gives me something to go by so I can verify further that I'm doing it right, and I can now check the relevant parts of the documentation. I find lot of time the documentation alone is not good enough if you don't know where to even look in first place.
Check out this thread on Gemini auto-research:
It's the first Reasoning Model to have access to the Internet:
Research in minutes:
x.com
x.com
Try Perplexity AI search engine:
Yeah if it can work with having real time access to internet that is incredible. ChatGPT is not continuously being trained and is based on a specific date, but imagine one that is, and a new version of software comes out and there's a common issue everyone is having, it would quickly learn on that and provide the relevant info.
They are at like 99.999% realistic with photos now. 2025 will be the Year of AI Catfishing lol
View attachment 114317
View attachment 114318
View attachment 114319
View attachment 114320
Now with motion. We can now trust 0% of what we see online: