Hacker News Reader

vincirufuson Diffusion based LLM basic chat app

very interesting where can i read more about Diffusion based LLMs, I did try asking dllm to explain itself and it did ok :-)

vincirufuson Vercel's x402-MCP Open protocol payments for MCP tools

this could be a game changer for the agentic era, similar to what online payments did to the ecomm & online banking era

vincirufuson ChatGPT is NOT a LLM – GPT is

Yup exactly!! its a subtle change happening under the hood, When people have a good interaction with ChatGPT they attribute it to the LLM, but in most cases, its the agentic features in ChatGPT that wows you and not necessarily the smartness of the large language model.

vincirufuson My productivity app is a never-ending .txt file

I stumbled upon this old post when researching on the best ways for my ai agent to maintain notes about our interactions. Applies to humans as well

vincirufuson ChatGPT is NOT a LLM – GPT is

Well there was is a lot of human element and infact this post was an output of conversations with different people talking about LLM and ChatGPT. I guess Claude over did the 'fix grammar and improve the language' instructions.

vincirufuson ChatGPT is NOT a LLM – GPT is

Sadly I don't have a Claude subscription for the number of tokens needed for Opus. Although I'd admit I used Sonnet to clean up and improve the language and grammar.

vincirufuson ChatGPT is NOT a LLM – GPT is

It used to be a thin wrapper, but todays chatGPT is closer to an agent

vincirufuson GLM 4.5 with Claude Code

update the title to not seem biased / hyped

vincirufuson GLM 4.5 with Claude Code

feedback accepted got rid of the killer bits

vincirufuson GLM 4.5 with Claude Code

Well I'd call them the poor person's claude code, wouldnt compare it with Opus but very close to Sonnet and Kimi

vincirufuson GLM 4.5 with Claude Code

Ahh bugger I pasted the wrong link I had this one open in another tab..

vincirufuson GLM 4.5 with Claude Code

yeah I too have heard similar concerns with Open models on OpenRouter, but haven't been able to verify it, as I don't use that a lot

vincirufuson There's still way too much confusion about OpenAI's Responses API

Yup. I think it was a nice honest and vulnerable thread

vincirufuson Next.js is infuriating

may the title of this should say Next.js middleware /logging is infuriating

vincirufuson AI is the natural next step in making computers more accessible and useful

There is already a growing number of people who use tools like super whisper to interact with agents. At work I'm building a "chief of staff" kind of a voice agent where you simply verbally give it tasks and it goes and gets stuff done. Also do have a look at the latest announcements from open AI on their realtime voice

vincirufuson AI is the natural next step in making computers more accessible and useful

I've been in tech and have been coding for 25 years now. My workflow in the past 2 months uses coding agents for 95% of my work. Just yesterday I had my agent fix and rewrite nearly 2 weeks of a 2 member in about 30 minutes. The agent's code was a lot lot better. There there is a huge bubble, but once that bursts new patterns will emerge, Just like how ecommerce and online banking emerged, after the doc com bust

vincirufuson AI is the natural next step in making computers more accessible and useful

Absolutely true!!!, there will always be a 1000 wrong ideas for 1 true revolution. Sometimes these pundits would be over ambitious in their thinking or off with their timelines. What MSFT said about voice is actually coming true now with AI Voice Agents as we speak..

vincirufuson AI is the natural next step in making computers more accessible and useful

How do you think people will write code and generate images 6 months to a year from now. Claude code, nano banana, perplexity, copilots, comet will become the default tools

vincirufuson AI is the natural next step in making computers more accessible and useful

What are your thoughts on the latest nano banana

vincirufuson AI is the natural next step in making computers more accessible and useful

Yup thats a valid way to look at it

vincirufuson AI is the natural next step in making computers more accessible and useful

This article was written a couple of months back look at the example of Photoshop and connect it to the nano banana release last week

vincirufuson [dead]

maybe thats just nature evolving, but surely the corals of today would be very different from the corals of a few decades before

vincirufuson The defense against slop and brainrot

Very well said, I have been telling people all around me that slop and brainrot will lead to the extension of humanity

vincirufuson Memento: Fine-tuning LLM Agents without Fine-tuning LLMs

Quite excited about this paper and how it could lead to building 'true agentic systems' that self learn

vincirufuson Writing Strong Content with AI

Interesting point on the monorepo. I've always suspected monorepos would also play a key role in helping coding agents digest large code bases, just havent been able to prove it.

vincirufuson How I got Claude-Code to work with a local LLM using a custom proxy

Kimi k2 works reasonably well for the price. Opus is still the king, but Kimi would be the poor persons's Sonnet. Gets the job done with a bit of extra help

vincirufuson How I got Claude-Code to work with a local LLM using a custom proxy

Neat I have a similar setup but using Kimi k2

vincirufuson Digg's new app is basic, but a great start

Wow is is Digg still around??

vincirufuson I tried DSPy and now I get why everyone won't shut up about it

bsolutely agree! Prompt engineering is brittle, hardcoded “templates” that don’t generalize, and it’s simply not scalable as tasks and use cases explode. DSPy’s approach, which builds reusable and composable reasoning patterns, is a huge leap forward. DSPy will become the default for serious LLM application development.