Nilesh Barla
@thenielspace
Researching AI Alignment and the limits of LLMs. Sharing my thoughts and findings from my research.
This week I was busy developing a custom GPT for research purposes. I aimed to create a foundational model that I experiment with on the go without requiring me to rent out GPUs. I wanted to see how my M1 Macbook Air would perform. Here are some insights:
We have come a long way from learning about the self-attention mechanism to understanding agentic LLMs.
Got up this morning and saw “Agent mode”. Will this make up for GPT-5 delay?
Like I said, AI or even a superintelligent AI will only be effective as we allow it to be. If we don't have enough domain knowledge, we will not be able to get anything good out of it.
I’m slowly beginning to accept that my productivity, when working with AI coding agents, is limited by my human brain. AI can do many tasks in parallel, but I can only track the context of a few, so I only run a few tasks at a time. I am the bottleneck.
“Superintelligent AI will help us develop holistic products." The superintelligent AI, or any current AI, for that matter, is as good as the questions you ask it or the task/prompt you give it. You will be more productive and result-oriented if you know what you want. That…
“Superintelligent AI will help us develop holistic products." The superintelligent AI, or any current AI, for that matter, is as good as the questions you ask it or the task/prompt you give it. You will be more productive and result-oriented if you know what you want. That…
“Superintelligent AI will help us develop holistic products." The superintelligent AI, or any current AI, for that matter, is as good as the questions you ask it or the task/prompt you give it. You will be more productive and result-oriented if you know what you want. That…
“Superintelligent AI will help us develop holistic products." The superintelligent AI, or any current AI, for that matter, is as good as the questions you ask it or the task/prompt you give it. You will be more productive and result-oriented if you know what you want. That…
GPT-5 getting final touches I see.
8/N Btw, we are releasing GPT-5 soon, and we’re excited for you to try it. But just to be clear: the IMO gold LLM is an experimental research model. We don’t plan to release anything with this level of math capability for several months.
We are almost there for GPT-5. Can’t wait!!!
GPT-5 Reasoning Alpha snapshot version 13-07 just spotted.
Have you solved the engineering problem leading to Superintelligence? Or are you closing in?
Today, we at @OpenAI achieved a milestone that many considered years away: gold medal-level performance on the 2025 IMO with a general reasoning LLM—under the same time limits as humans, without tools. As remarkable as that sounds, it’s even more significant than the headline 🧵
Now I am curious about product alignment. I mean, if we are not aligning our products with user preferences, then: 1. We might be using AI wrong. 2. AI is not yet capable of providing us with proper feedback from user data or behaviour. 3. Or worse, we are not clear about the…