sagarpatil

I agree with the author. What’s recently worked for me is to ask the AI to critique my blog/hn/reddit posts and ask it to be brutally honest. The feedback that I got from it was eye opening and I used to write exactly like the author mentioned: a lot of meandering.

I’ve been playing with Gemini 2.5 pro throwing all kinds of problems that will help me with personal productivity and it’s mostly one shoting them. I’m still in disbelief tbh. A lot of people who don’t understand how to use LLM effectively will be at an economic disadvantage.

Could use multiple Google accounts to increase the rate limit.

Yeah, if I don’t have higher rate limits, it’s useless. This just sounds like a gimmick launch where they want to gather feedback. It will be a couple of months before this will be GA.

Any idea what is RPM for this model?

My prompt is nowhere near yours.

Just for fun: Asked it to rewrite the first page of ‘The Fountainhead’ where Howard is a computer engineer, the rewrite is hilarious lol.

https://gist.github.com/sagarspatil/e0b5443132501a3596c3a9a2...

AI labs, it seems, use a template for system cards as well. OpenAI stands out because they showcase their employees using their tools for various use cases, which is refreshing.

The only benchmark I care about. Thanks!

Couldn’t have said it better. I’ve developed multiple agents with just function calling, structured outputs and they have been in production for more than a year (back in the day we did not call it agent lol) I think these is targeted towards people who are already using agent frameworks + OpenAI API.

Say you hire a developer and ask him to directly debug an issue by simply skimming through the codebase, do you think he can complete this task say in 5-10 minutes? No, right? In claude code(CC), do the following: 1. /init which acts as a project guide. 2. Ask it to summarize the project as save it as summary.md 3. The prompt needs to be clear and detailed. Here’s an example: https://imgur.com/a/RJyp3f9

I have one word: Woow!

The fundamental issue with these AI safety declarations is that they completely ignore game theory. The technology has already proliferated (see: DeepSeek, Qwen) and trying to control it through international agreements is like trying to control cryptography in the 90s.

I've spent enough time building with these models to see their transformative potential. The productivity gains aren't marginal - they're exponential. And this is just with current-gen models.

China's approach is particularly telling. While they lack the massive compute infrastructure of US tech giants, their research output is impressive. Their models may be smaller, but they're remarkably efficient. Look at DeepSeek's performance-to-parameter ratio.

The upside potential is simply too large to ignore. We're seeing breakthroughs in protein folding that would take traditional methods decades. Education is being personalized at scale. The translation capabilities alone are revolutionary.

The reality is that AI development will continue regardless of these declarations. The optimal strategy isn't to slow down - it's to maintain the lead while developing safety measures in parallel. Everything else is just security theater.

(And yes, I've read the usual arguments about x-risk. The bottleneck isn't safety frameworks - it's compute and data quality.)

Cool story bro. Genuinely.

Probably the best advice so far. Job = limited upside. I’ve seen a lot of normal people start business and become financially stable.

Sure you are kidding Mr. Numba88

I’ll probably get down voted to oblivion, but this is really silly. I do not see any value proposition. To each their own, I guess.

I’ll do it. AFL right now.

Just 20 light years away (As a normie I giggled a little).

Here’s why. Because you can chain the prompts, CoT and answers. Let me explain.

Prompt 1 (64k) CoT (32k) Answer 1 (8k)

CoT 32k context is not included in the 64k input. So it’s actually 64k + 32k + 8k.

Prompt 2 (32k) + Previous CoT 1 (32k - this time it will be counted because we are chaining and these are two different API calls) Answer 2 (8k)

Another way to optimize this is to use another model to pick up only the correct CoT from the current answer and pass that as CoT for the next prompt. (If you are feeling adventurous enough, you could just use R1 to select the correct CoT but I think it will go insane trying to figure out the previous and current CoT)

I just use cursor agent to do git now (unless it’s simple stuff)

A glorified reminder? Really?

It’s mind blowing how we went from Adobe Flash to this in just 25 years.

msty.app does this and much more. It’s open source too.

As someone who recently moved from a mechanical keyboard to apple keyboard, the closest thing that can imitate the mechanical key sound is through this macos app: https://tryklack.com/

My only gripe with mechanical keyboards is there’s no version with apple touch ID (I know the technical constraints).

You can use firecrawl.dev to get markdown content for any page you want. It’s open source and you can host it locally.

Wow. I was thinking of implementing spaced repetition for a project I’m working on, this will be very useful. Thank you for creating it.

2 million token size is not enough?

$99/month lol. I have Perplexity, OpenAI, Claude and Cursor subscription and I end up paying way less than $99/month. Clearly you haven't done any research on price. Aider, Cline are open source, in not sure why someone would subscribe to it unless it's the top model on http://swebench.com/

Been using Cursor since launch. Really frustrating how they charge per message (500/mo) instead of by token usage. Like, why should a one-line code suggestion cost the same as refactoring an entire class? Plus it's been losing context lately and making rookie mistakes.

Tried Zed AI but couldn't get into it - just doesn't feel as smooth as Cursor used to. GitHub Copilot is even worse. Feels like they're so obsessed with keeping that $10/month price point that they've gutted everything. Context window is tiny and the responses are barebones.