dangerface

Joined 12 years ago

Karma
1,535
Submissions
921
Member since
May 20
View on Hacker News

Recent submissions

Comment · 3 hours ago
Running Qwen 3.8 27B on a 5090, 8bit quant and 160k context fills the card but I get about 200 tps. Not as good as claude but its the first local model I have tried where it could easily replace claude. The local setup takes about twice as long as claude to complete a task to the same output quality. Electricity is expensive here so running a 600w gpu 24/7 ends up costing about the same as just buying credits. Free heating for winter and no spying are still worth it.

I went for the 5090 for the software support but I have so many issues running cuda on fedora. Kinda wish I had bought a Nvidia spark or something that was an AI only machine.

Comment · 7 days ago
I learned css from a book back in the 90s. In the 2000s tutorial sites like this where amazing, I thought people where crazy to buy a book and learn the way I did. Today ChatGPT is amazing for learning, why would any one still use tutorials?

This is like buying the IRC channel #coding a few years after stack overflow has taken over. What am I missing here?

Comment · 7 days ago
Started a job a two weeks ago with lots of php 5 websites that can't be updated because they are abandonware and management don't want to fork out for a replacement.

I used httrack to create a static version of the site and thats whats public now. Has bought me time to talk with management and convince them of what needs doing.

Comment · 14 days ago
Built a text based adventure. It works like D&D but with an agent as the GM. Feed it in campaigns and it builds a new one for you to play.
Comment · 14 days ago
I gave openclaw a bitcoin wallet and $100 to spend. I told it the $100 was our marketing budget for the month and it should come up with a marketing strategy and then implement it using the budget.

With in minutes it had spent all the budget buying fake follows on an X account that was similar to mine.

Lessons learnt: Most checkout pages have strong anti bot / fraud protection it struggles to get past. Most checkouts require an account which it struggles to register. It can't budget, if you give it money to spend, it will do as you ask and go to work spending it. It doesn't really know what it needs. It wanted to buy a vps despite already having access to a large dedicated server. It doesn't understand value. It wanted to buy the most expensive package because it offered the most features, it through that means its the best value.

In my case I gave it an email account it could sign up with and then used the bitcoin cli to make the payment.

It had its own wallet so it couldn't spend more than what was in it. I didn't want to limit it to for example $10 a day because I wanted it to be able to buy in bulk. Most advert space will be like $70 for a week not $10 a day. So I thought a daily limit would not work unless I put a large budget behind it.

It misses the entire point of money tbh. It just doesn't understand it.

Creating accounts and fraud protection are the biggest issues tbh. We need an alternative to captcha. Something that protects registration and checkout pages from abuse but can still be solved by agents. If you solve that issue the rest is already solved, just give it a credit card or crypto account. Most corporate cc providers allow you to put limits and alerts on the cards.

In a few years that will be the norm but for now I think agent marketplaces are the best solution. Instead of trying to work around bot protection on ebay just make an agent friendly ebay. That's what I am building.

Comment · 1 year ago
Yup even when you tell it your version it forgets pretty quickly. Or agrees it messed up and assures you this time it will give you the correct info for your version number then gives you the same command.

Javascript is a nightmare as they change everything constantly. PHP has backwards compatibility for everything so its not really an issue.

It also gives out dated info on salesforce, and im not just talking about the latest and greatest, it recommends stuff that was deprecated years ago.

Comment · 1 year ago
ZSA Moonlander is an easy avoid. They have a defect on their thumb cluster, the plastic that holds them to the main board is thin and breaks after a few years of use. Good thing they sell spares so they can be easily repaired but bad thing is trying to order them. They don't want just any one buying parts and even sending a photo proof you own a broken Moonlander is not acceptable, they want a receipt of the original purchase or they refuse to sell them.
Comment · 2 years ago
Since you don't use IRC for anything more than an unreliable and easily spoofable key value store, why bother with implementing an IRC client? You could easily write a client server on UDP 58 client sends the key and server replies the value. Create a list of 13 IP we just trust and run your dnssec on top of that? Ez
Comment · 2 years ago
You claim here that there is a cost to creating the domain but rule out Ens because it has a cost? You can't make both claims. You can generate Ethereum and buy domains with it much in the same way you do here.

You rule out DNS as it doesn't have a standard registration but IRC doesn't have a standard way to register channels nickserv isn't on all IRC and they have different implementations.

IRC is not distributed it's owned by a centralised provider who has complete control, they can trivially take over any domain they want.

IRC has no consensus mechanisms a netsplit would make it trivial to create duplicate channels on the same IRC network.

You check channel availability then register, this by definition isn't atomic, it's a race condition between check and register. You need a way to wrap this in a transaction to prevent the race condition. IRC provides no way to do transactions. Even registering a channel on a network is not atomic because of the potential of netsplits, which is stated in the rfc.

How are networks choosen and distributed? You picked 11 random servers that meet your requirements. If I pick 11 random servers how are their ips shared? Traditionally this is done with DNS but since you are trying to replace this requiring it is odd, how do you stop some one just hijacking your DNS once that problem is solved your scheme becomes pointless.

If you expect everyone to use the same 11 servers this isn't distributed, it's centralised to 11 servers you can't trust.

I think you are under the impression that IRC is distributed and so if you build a secure system on top of it, then the system as a whole will be secure and distributed. The problem is that IRC isn't secure or distributed it's centralised with no Byzantine fault tolerance.

I understand you check multiple channels/networks in an attempt to provide this protection but the underlying servers can do anything they want and there is nothing you can do about that. Finally getting to the point of being on IRC and checking a channel has a lot of attack points none of which are accounted for.

The difficult part of any distributed bysentain tolerant system is the distributed bysentain problem IRC doesn't solve this problem and your scheme outsources it to IRC.

Charcircuit is just telling you the truth you don't want to accept.

Comment · 2 years ago
They will keep selling cars but put more focus on their AI capabilities to stay on trend. I think we can already see this with their llm and robots, once they start making their own chips Elon will want vertical integration and start making their own silicone. the focus will be more on the autonomy of the car and larger tesla ecosystem rather than on the fact the cars are electric since every car is now electric.
Comment · 2 years ago
This is very true its the unintended consequences of engineering that cause the most harm and are most often covered up. I always think of the example of the hand dryer that can't detect black peoples hands and how easy it is for a non racist engineer to make a racism machine. AI safety putting its focus on the what if it decides to do a genocide is kind of silly, its like worrying about nukes while you give out assault riffles and napalm to kids.
Comment · 2 years ago
Strangely I think Ilya comes out of this well. He made a decision based on his values and what he believed was the best decision for AI safety. After seeing the outcome of that decision he changed his mind and owned that. He must have known it would result in the internet ridiculing him for flip flopping, but acted in what he thought was the best interest for the employees signing the letter. His actions are wroth criticism but I think his moral character has been demonstrated.

The other members of the board seemed to make their decision based on more personal reasons that seems to fit with Adams conflict of interest. They refused to communicate and only now accept any sort of responsibility for their actions and lack of plan.

Honestly Ilya is the only one of the 4 I would actually want still on the board. I think we need people who are willing to change direction based on new information especially in leadership positions despite it being messy, the world is messy.

Comment · 2 years ago
Its not a private company it is a non profit working in the public interest this usually requires some sort of public accountability. The board want to be a public good when they make decisions but want to be a private entity when those decisions are criticised by the public.
Comment · 2 years ago
Yea its a bit much he obviously doesn't deserve the admiration that he is getting. That said he deserves respect for helping bring ChatGPT to market, he deserves support because the board have acted like clowns and justified it with their mission of public accountability, but have rejected the idea that the board itself should be publicly accountable.
Comment · 2 years ago
Keeping D'Angelo on the board is an obvious mistake, he has too much conflicting interest to be level headed and has demonstrated that. The only people that benefited from all this are Microsoft and D'Angelo. Give it a year and we will see part 2 of all this.

Further where is the public accountability? I thought the board was to act in the interests of the public but they haven't communicated anything. Are we all just supposed to pretend this never happend and that the board will now act in the public interest?

We need regulations to hold these boards which hold so much power accountable to the public. No reasonable AI regulations can be made until the public are included in a meaningful way, anyone that pushes for regulations without the public is just trying to control the industry and establish a monopoly.

Live from the official Hacker News API

A modern reader for Hacker News · SvelteKit + Skeleton