HNHacker News
TopNewBestAskShowJobs

yousif_123123

415 karma · joined June 20, 2023

submissionscomments
yousif_123123··on GPT-5.2
Why doesn't OpenAI include comparisons to other models anymore?
yousif_123123··on Fara-7B: An efficient agentic model for computer use
Perhaps they want to be able to run them on mobile hardware they release?
yousif_123123··on Maybe you’re not trying
I was hoping someone points it out for us.

Since you asked me, you are using the same concept and now I need to help you solve your problem (which seem to be the one I also have..)

I think the solution must be we're primarily responsible for ourselves, and that unless we ask others for help all the time we need to figure things out. I also lately have been thinking from the perspective of the person I'm anxious to interact with, and feel that they may actually be happy to interact with me, receive some warm greeting and help out by answering my question or doing my task.

If you could do something for others but feel anxious doing it for yourself, it must be "in our head" and logically we should be able to get over that and choose to be brave. I think in really it's often missed how we can be brave doing the action if it was for someone else, and that the bravery may actually already be inside us.

This at least is how I think of it now.

yousif_123123··on Maybe you’re not trying
I've always noticed that when I'm giving advice to someone or trying to help out, it always feels their problem is easier than whatever problem I have. As someone with some anxiety around things like calling some company to get something done or asking a random stranger for some help in a store, I would gladly do it if it was to help someone else (family member or friend). But when it's for me I find it harder.

I wonder how much psychologically we can be more confident and less anxious when we're doing something for others vs ourselves..

yousif_123123··on WriterdeckOS
Probably to help you avoid being distracted given the higher friction of rebooting if your dual booting or going to another device vs just launching another browser tab..
yousif_123123··on Dating: A mysterious constellation of facts
Happy in terms of being a customer of the app perhaps.
yousif_123123··on The next chapter of the Microsoft–OpenAI partnership
Tether is projected to generate $15 billion in profits. So 500 billion is like a 33 times earnings multiple.

Now the main thing is how sustainable these earnings are and if they will continue to be a dominant player in stable coins and if there will continue to be demand for them.

Another difference to Blackstone is Tether takes 100% of the returns on the treasuries backing the coins, whereas Blackstone gets a small fee from AUM, and their goal is to make money for their investor clients.

If crypto wanted to really be decentralized they'd find a way to have stable coins backed by whatever assets where the returns of the assets still came to the stable coin holder, not some big centralized company.

yousif_123123··on The next chapter of the Microsoft–OpenAI partnership
I think I read that its valued at 500 billion based on their latest fund raise. I don't know the total holdings they have.

I have no interest in crypto, just wanted to mention this which was surprising to me when I heard it.

yousif_123123··on The next chapter of the Microsoft–OpenAI partnership
I think the stable coin company Tether is valued at 500 billion also.
yousif_123123··on AMD signs AI chip-supply deal with OpenAI, gives it option to take a 10% stake
I think at least part of the 10% is if AMD stock reaches 600.

Not that I disagree that this looks weird. Why was that needed to be offered? Couldn't they just buy the AMD chips if they're good enough? Or Nvidia is it's better?

I also don't get why there commiting so much to the future, are they sure of the quality of the products and their demand that much?

yousif_123123··on AbsenceBench: Language models can't tell what's missing
Maybe if instructed, but how would it know it needs to use python in this case vs just answer? Perhaps you'd instruct it to always attempt using code to reduce errors.

But the idea of trivial problems like this potentially causing issues for LLMs might mean other aspects of intelligence could also be a struggle for LLMs (which could impact it's coding ability as well).

yousif_123123··on AbsenceBench: Language models can't tell what's missing
This is very interesting. 1. Authors mention the attention mechanism being perhaps unable to attend to the location of gaps since the gaps aren't tokens. But I would've expected a good LLM transformer to be at least a bit close to the gap location. I don't understand why mathematically the architecture is less suitable for that, it could attend to a region that may contain gaps. I wonder if fine-tuning on a task like this could help? 2. Shorter inputs with less omissions were harder to solve. That is not completely surprising, as a human doing this task, if 1 word was missing it would be harder to notice. And similarly 1 line would be harder than 10 lines. But still interesting for an LLM to have this problem. 3. Reasoning models do better, as they can write out the documents and potentially solve this easily. It still very surprising that this doesn't lead to 100% accuracy. This should be a trivial task. Like the paper says, a trivial program can be written to solve this. Perhaps ChatGPT (or similar agent) could read this paper while training, and know to write and run python when solving an issue like this.

The most interesting thing though, is what other aspects of intelligence we may not have identified explicitly, and whether LLMs and current AI is very bad at them. This paper suggests that there likely are many of those, and it seems in general a pretty fun time for people working building benchmarks.

yousif_123123··on VPN providers in France ordered to block pirate sports IPTV
Also just more accessibility in terms of being able to pay for it. As a viewer in Canada, there was no service whatsoever that was showing the Real Madrid vs Barcelona Spanish cup final a few months ago (Copa del Rey). I had to signup on a Spanish state tv website and use a VPN to access it for free.

I would have gladly paid, but there was no opportunity.

Content fragmentation and some sports rights not being bought and resold by anyone is also a big problem.

yousif_123123··on The Art of Fugue – Contrapunctus I (2021)
The Emerson quartet version blew me away when I first heard it. I recommend it over the piano versions because I think you hear the full extent of the voices better (coming from someone that loves and play the piano).
yousif_123123··on Gemini 2.5 Pro Preview
The opposite problem is also true. I was using it to edit code I had that was calling the new openai image API, which is slightly different from the dalle API. But Gemini was consistently "fixing" the OpenAI call even when I explained clearly not to do that since I'm using a new API design etc. Claude wasn't having that issue.

The models are very impressive. But issues like these still make me feel they are still more pattern matching (although there's also some magic, don't get me wrong) but not fully reasoning over everything correctly like you'd expect of a typical human reasoner.

yousif_123123··on OpenAI releases image generation in the API
Yes. Also, when testing low vs high, it seems the difference is mainly in the diffusion part, as the structure of the image and the instruction following ability is usually the same.

Still, very exciting and for the future as well. It's still pretty expensive and slow. But moving in the right direction.

yousif_123123··on OpenAI releases image generation in the API
It's also good but clearly not close still. Maybe Gemini 2.5 or 3 will have better image gen.
yousif_123123··on OpenAI releases image generation in the API
It's a mix of both it feels to me as I've been testing it. For example, you can't get it to make a clock showing custom time like 3:30, or someone writing with their left hand.. And it can't do follow many instructions or do them very precisely. But it shows that this kind of architecture will be be capable of that if scaled up most likely.
yousif_123123··on Gemini 2.5 Flash
I have noticed Gemini not accepting an instruction to "leave all other code the same but just modify this part" on a code that included use of an alpha API with a different interface than what Gemini knows is the correct current API. No matter how I promoted 2.5 pro, I couldn't get it to respect my use of the alpha API, it would just think I must be wrong.

So I think patterns from the training data are still overriding some actual logic/intelligence in the model. Or the Google assistant fine-tuning is messing it up.

yousif_123123··on GPT-4.1 in the API
I do like the vinyl and analog amplifiers. I certainly hear the warmth in this case.
yousif_123123··on OpenAI Audio Models
Well I think there's many cheaper models in terms of bang for buck currently per token and intelligence than gpt4o. Other than OpenAI having very high rate limits and throughout available without a contract done with sales, I don't see much reason to use it currently instead of sonnet 3.5 or 3.7, or Google's Flash 2.0

Perhaps their training cost and their current inference cost is higher, but what you get as a customer is a more expensive product for what it is, IMO.

yousif_123123··on GPT-4.5
It's disappointing not to see comparisons to Sonnet 3.7. Also since o3-mini is ahead of o1, not sure why in the video they compared to o1.

gpt4 was way ahead of 3.5 when it came out. It's unfortunate that the first major gpt release since that is so underwhelming..

yousif_123123··on DOOM CAPTCHA
It does have it. If you tap on it, it should show it.
yousif_123123··on Artificial needles to real haystacks: Improving retrieval capabilities in LLMs
Haven't read the paper yet, but looks like this can improve the ability of the model attention to work better, since many of these tasks end up being similar to these generic tasks.

Even gpt4 gets tripped up when there's too many exact instructions needed to be executed on an input. That's why it's common that breaking a task into multiple steps and multiple improves performance.

It's wonderful to see improvements possible on smaller models.

yousif_123123··on Testing Generative AI for Circuit Board Design
One downside for diffusion based systems (and I'm very noob in this) is that the model won't be able to see it's input and output in the same space, therefore wouldn't be able to do follow-up instructions to fix things or improve on it. Where as an LLM generating html could follow instructions to modify it as well. It's input and output are the same format.
yousif_123123··on AI Search: The Bitter-Er Lesson
Still, similar to when you have read 10 textbooks, if you are answering a question and have access to the source material, it can help you in your answer.
yousif_123123··on Cost of self hosting Llama-3 8B-Instruct
deepinfra.com hosts Llama 3 8b for 8 cents per 1m tokens. I'm not sure it's the cheapest but it's pretty cheap. There may be even cheaper options.

(Haven't used it in production, thinking to use it for side projects).

yousif_123123··on ChatGPT Edu
Yes it could be like learning how to use google to get information, or to learn the logic to solve something. I think the main under hyped aspect of AI currently is how useful it can be integrated with whole bunch of regular programming code and UI/UX to implement parts of functionality that can't be programmed explicitly. Like even if gpt4 is the last AI model, if it keeps getting cheaper and faster, the amount of cool features that can be with with some code and a few API calls in parallel or in sequence will add a lot of nice features to all the apps people interact with everyday.

Like my Dad uses multiple apps to get the information he wants for all the sports games that happen during the week, and no app is perfect. If he could prompt something to make a UI with the relevant info he wants displayed like he wants, he would be so much happier. It's little things like that that will make software way more customized and useful for people.

I think there's a lot of hype on AGI and they'll be more, but just as another tool in the programmer's toolbox, LLMs will allow better product to be developed (if people learn how to prompt and think with LLMs in mind vs dismissing all of it when they see a mistake)..

yousif_123123··on ChatGPT Edu
At the same time using chatgpt or any LLM productively, and getting the most out of it to help you think and reason better is a skill on its own that requires development.
yousif_123123··on The deskilling of web dev is harming us all
I don't fully disagree, but this might be subjective.

For me as a full stack developer working with small teams/startup, I actually don't consider myself full stack. I just want to be able to do whatever it takes to make a product and ship features. Does it need a websocket server? I'll learn how to do that. Does it need advanced client side caching? I can do that, etc.

To some extent "product development" is both art and engineering. In the engineering side, you can think of html and CSS and http and testing as different things requiring a multidisciplinary team.. but if you think just in terms of "building the thing", I like to feel that I can get it done with whatever technology needed. That's why I got into programming in the first place. Not to write code and be an "engineer", but to make the computer do cool things.

AI does expand the capabilities of someone that wants to get things done. I have written in languages that I don't have experience with, and recently was exploring a neo4j db with cypher queries written with ChatGPT (I have only MongoDB experience), something that would've taken me hours to learn.

Still, having experts in specific areas in a team can be very helpful (both to get technically difficult things done but also to have others learn from them).

I just don't want to be 1 part of a team specializing in my limited domain, where I start to care more about the technical part than the actual value delivered or user experience..

I think what might cause me to burnout is too much specializing, too much bureaucracy, people telling me I can't do this, we need to hire a staff level person to this thing etc..

← PreviousPage 2 of 3Next →