HNHacker News
TopNewBestAskShowJobs

segmondy

9,039 karma · joined January 30, 2013

Builder of thingz, lover of old robust tech like unix, prolog, lisp, APL, forth, postgresql, but don't mind the shiny new things ... right now, I'm all in on GenAI, Local LLMs & LLM driven agents, team llama.cpp, building LLM agents since 2023

segmond AT gmail dot com

submissionscomments
segmondy··on Source code of the Estonian online voting system
I think this is an easy enough problem to solve, the issue is that someone else can how force you to reveal who you voted for. Spouse, rouge employers, mafia, corrupt politicians. You can now "sell" your vote.

I think such a system could still work, but just like the voter has to go to the booth alone to vote, they must go the booth alone to verify.

segmondy··on AI Revolution vs. Computing Revolution
We don't know, predicting the future is hard. What we know is that some companies around today will not be around in the future, and some new companies will pop up. Between now and then, a lot of money will be made by some people and a lot will be lost by others.
segmondy··on Ask HN: For those who left Big Tech for early-stage startups, do you regret it?
Well, I stayed at a big tech instead of going to an early-stage startup after I got burned in a startup. I knew the founder of this startup, but I needed money I can touch not paper money, they exited and everyone that was there when I could have joined have been handsomely retired for a few years now while I sweat about job stability and retirement.
segmondy··on Ask HN: Thoughts on hiring junior engineers in 2026?
It's not about junior vs senior. it's about hiring a competent developer. There are competent junior developers and there's plenty of useless seniors. Besides these, what you are building matters. If you are not dealing with hard tech, then a junior with a model could level up fast. If you are building complex hard tech then experience matters a lot for judgement, however some juniors could still keep up. The bar is at an all time high, but for everyone, seniors as well.
segmondy··on I know why the mathematicians weep
I have no idea why they weep, I read the article and I believe they weep because they choose a mindset of scarcity vs abundance. There's going to be more math with AI! The number of math to solve is infinite, AI will bring to the surface so many of them and can't solve all of them, a person can pair up with AI to tackle one.
segmondy··on Ask HN: What is the black swan event that could bring down GPU costs?
China
segmondy··on Muse Spark Open
AA is a joke.
segmondy··on Is 'Attention' the Essence of Intelligence?
Not to attack you, but even a modern model will make a more coherent argument than this.
segmondy··on TV4 postpones show after AI agent erases edited material
what do you need backup for, when you regen everything in 5 seconds with AI. perhaps they replaced backup with AI too.
segmondy··on Intel D270 GPU AI Accelerator–Up to 480 GB LPDDR5X
you won't be able to buy it for $200 in 7 years. 7 year old GPUs are still going for more than that, and take into account inflation and demand for these things. My bet is that you will not be able to buy it not even for < $4000 in 7 years
segmondy··on Solar panels that plug into your wall
Once this becomes cheap enough and easy enough, I expect heavy handed regulation pushed by the power companies. Chances are we will buy these, plug it in and our energy bills will be the same or more!
segmondy··on 'AI' Is Not a Tool by Rich Hickey
I'm a greybeard too and by that post they made, they are completely out of touch.
segmondy··on AI models can provably hide arbitrary capabilities
no it's not. LLMs just output the text, they don't execute command. You have to capture that text, then parse it and decide what to do with it. You can for instance execute it blindly. Or before execution you can inspect it and decide if it's safe or not. You can now take a decision model like jev and have it inspect the command to see if it's safe or not before execution. If your LLM model and decision model is not coming from the same level, you have little to worry about.

But let's put aside decision models, you should be running your agent in a sandbox.

segmondy··on 'AI' Is Not a Tool by Rich Hickey
Your heroes could be wrong, I'm a fan of all these folks. But this time around, not on their side. Guess who is on my side? Linus Torvalds for one. see - https://www.phoronix.com/news/Linus-Torvalds-Claude-Code
segmondy··on 'AI' Is Not a Tool by Rich Hickey
you should see what Linus Torvalds is saying about LLMs
segmondy··on 'AI' Is Not a Tool by Rich Hickey
So it just needs to have one of those properties and not all? LLM is a tool, IMO, it's pretty simple to use it too.
segmondy··on 'AI' Is Not a Tool by Rich Hickey
possibly, i drank the kool aid 3 years ago, when the context window was 4k, i never put the cup down.
segmondy··on Just say no to OpenAI and Anthropic
Running local is not as difficult as it is, if the will is there and you are resourceful, you can do it.
segmondy··on I am so tired of startups doing worse than me
Hah, your mistake is thinking that startup is about building tech. Maybe years ago. Startup is about building a startup, your goal is to do what you need to do to get funded and that is often not building an amazing tech. The fancier your tech the more you come across like a hardcore nerd with no business acumen. This is what the capital owners often think when they come across your type. If you were so smart, why would you waste all that effort building it instead of raising money or getting other people to build it? If your tech is so great, why don't you have bazillion users? Hope you begin to see what you are doing wrong, stop building and optimize for raising money, not users, not more features.
segmondy··on The Hetzner Cloud network stack – history and technical overview
find an alternative, no one is forcing you to use them. i never understand why people complain about pricing if they have a better choice.
segmondy··on You're saving one minute and now I can't know if I'm talking a human
Use AI to reply? Use AI to read the message, pass it to Pangram. If Pangram detects it's AI, your AI will reply. Since your input was generated with AI, my AI is replying in kind. Have an AI inbox/label, everything AI goes in there.
segmondy··on I think I found a planet nobody knew existed. I used Claude Code to find it
some of us like to go for solo walks all alone.
segmondy··on Docker Agent
Better read the terms and conditions and if they can retain your data, docker now collects inputs and outputs ...
segmondy··on Ask HN: Which frontier model can do code security reviews
all of them
segmondy··on Mistral Large 4
I use Opus 5.5 at work.

I use MiMov2.6Pro, DeepSeekv4.1Flash, GLM5.3, Hy4, Qwen3.8 and KimiK3 at home. Opus5.5 is not a game changer.

segmondy··on Mistral Large 4
I tried Mistral's last 3 large models, devstral and mistrallarge3 and the numbers were not even benchmaxxed, but just false. the models were so weak and garbage. Let's hope they are telling the truth this time around, we need more alternatives. There mistral-small and original MistralLarge models were awesome, hoping they are back!
segmondy··on Beam: Reflection's 501B open-weight model
Any time a new lab shows up, folks complain about how their models are worse. Really? It would be nice if a new comer comes from no where and beats everyone, but that's rarely the case. The good thing is that other labs/people are figuring out how to build this, and if they keep at it then this is as bad as it gets for them and it would hopefully get better. A new entrant to the market is good for everyone.
segmondy··on Run Qwen 3.8 Flash Next (125B) on consumer hardware (RTX 4090) at 100T/s
ds4 is the inference engine not the model.
segmondy··on Run Qwen 3.8 Flash Next (125B) on consumer hardware (RTX 4090) at 100T/s
The larger the model, the more you can go down. K3 in Q1 will match and likely beat Qwen3.8-Flash-Next.
segmondy··on Docker has always used microVMs (well since 2016)
I was just exploring the new docker sandbox and saw that they could retain all your data in/out of the sandbox for 30 days. I wash my hand of docker, use with care. Data is gold and lots of companies are now going to be giving their "free" products as a sort of trojan horse to steal your data.
← PreviousPage 3 of 34Next →