HNHacker News
TopNewBestAskShowJobs

mynti

338 karma · joined July 23, 2024

submissionscomments
mynti··on Ilya Sutskever: We're moving from the age of scaling to the age of research
If we think of every generation as a compression step of some form of information into our DNA and early humans existed for ~1.000.000 years and a generation is happening ~20years on average, then we have only ~50.000 compression steps to today. Of course, we have genes from both parents so they is some overlap from others, but especially in the early days the pool of other humans was small. So that still does not look like it is on the order of magnitude anywhere close to modern machine learning. Sure, early humans had already a lot of information in their DNA but still
mynti··on Estimating AI productivity gains from Claude conversations
".. Claude estimates that AI reduces task completion time by 80%. We use Claude to evaluate anonymized Claude.ai transcripts to estimate the productivity impact of AI."

What is this? So they take Claude and ask how much do you think you saved on time here? How can you take this seriously. ChatBots are easy to exaggerate, especially about something positive like this.

mynti··on Altman's eye-scanning startup told workers not to care about anything but work
We are so much more productive and efficient, so we should even work on weekends to be even more productive.. and then they make an ID System that has been solved for ages with simple passports
mynti··on Gemini 3 Pro Model Card [pdf]
It is interesting that the Gemini 3 beats every other model on these benchmarks, mostly by a wide margin, but not on SWE Bench. Sonnet is still king here and all three look to be basically on the same level. Kind of wild to see them hit such a wall when it comes to agentic coding
mynti··on GPT-5.1: A smarter, more conversational ChatGPT
So after all those people killed themselves while chatgpt encouraged them they make their model, yet again, more 'conversational'. It is hard to believe how you could justify this.
mynti··on Show HN: Myna - monospace typeface for symbol-heavy programming languages
Thanks for creating this! I have been trying it out and it looks like it is a more dense monospace than my other ones, so I can actually see more horizontally on my screen.
mynti··on Making GPT-2 better at math reasoning with a new attention mechanism
Cool idea! I had a look at the code and have been wondering about the sigmoid gating, it is used to add some of the q_struct and k_struct into the original key and query. But I wonder why this gating is independend of the input? I would have expected this gating to be dependednd on the input, so if the model sees something more complex it needs more of this information (or something similar). But it is just a fix, learnable parameter per layer, or am I mistaken? What is the intuition about this?
mynti··on Annual hours worked per worker in OECD countries
If you put it against the value created from these hours, the graph almost flips entirely: https://figure.nz/chart/mMmSnWWbULiK4SvY-17BBScq4PaYeiUnz

Also: in some countries, like Germany, there is a lot of part time work for mothers, which does impact this statistic quite a bit

mynti··on Experimental 3.5T K2 merge beats GPT-4.5 and Opus at writing
"beats at writing" is kind of far fetched since the page provides three generic pormpts with answers from Opus, GPT5 and K3. But a 3.5T model is crazy huge and the answers still feel more or less the same to the other models
mynti··on The Continual Learning Problem
Super interesting blogpost. I just wonder how this is actually different to LORA, since LORA also adds some parameters and freezes the rest of the model. This seems like a sparse, memory efficient LORA with a couple of extra steps, since it uses attention again to make the sparsity work. All while making it a lot more effective compared to LORA (performance drop of only 11% compared to 71%).
mynti··on Optimizing Your Content for Inclusion in AI Search Answers
Make others turn the web into a perfect LLM training set, perfect strategy
mynti··on The cost of turning down wind turbines in Britain
This will probably take a little longer for private use, but the industrial sector is already doing this. Cooling chambers being cooled down further during cheap electricity prices (or sunshine when they have their own solar) or storing heat/"cool" underground
mynti··on Tether CEO confirms major capital raise at a reported $500B valuation
For reference: Circle (USDC) has about 74B market cap and is valued at 30B. Tether (USDT) has about 172B market cap and is about to be valued at 500B. I get that Tether is the market leader and all, but that valuation is totally nuts
mynti··on LongCat-Flash-Thinking, LLM from Meituan (China's Equivalent of Uber Eats)
No clue why they are calling a 560B model "Flash". But one very interesting thing is, it is beating all the other frontier models in safety, and most of them by quite a margin. The same for Formal Theorem Proving, where I would have expected OpenAI to do way better
mynti··on Everyone's trying vectors and graphs for AI memory. We went back to SQL
How does Memori choose what part of past conversations is relevant to the current conversation? Is there some maximum amount of memory it can feasibly handle before it will spam the context with irrelevant "memories"?
mynti··on The cost of performing childhood for your parent's art
“To be an artist means you must declare a loyalty to your art form and your vision that runs deeper than almost any other, even sometimes deeper than blood kinship.”

That is absurd. Selling out your childrens childhood for some sort of pseudo deep "art" and defending it in this way is almost psychopathic

mynti··on How big a solar battery do I need to store all my home's electricity?
Balcony solar is absolute awesome in germany. I get about 30% return on investment per year on my small solar panel. Hard not to do it. I have no idea why it is still a little niche
mynti··on Qwen3-Next
For anyone curious about what the Gated Delta Network is: https://arxiv.org/pdf/2412.06464
mynti··on A wild week: Grok Code Fast 1 exploding to 66% usage share
This is a bit disingenous since the tokens are free until September 10th, so of course the usage is spiking and people are flocking towards it.
mynti··on Benchmark for Local LLMs with German "Who Wants to Be a Millionaire" Questions
This is super cool! One thing I find counterintuitive is that GPT5 or o3 not have better performance. GPT5 gets about 800k on average per round but I would have expected it to be nearly perfect, since these are not particularly hard questions and mostly trivia or simple look up knowledge questions. There is little reasoning involved so I expected the big models to do much better.
mynti··on Qwant and Ecosia launch Staan, their search index as an alternative to Google
i have been anticipating this release for some time now. I am an avid user of ecosia and find the results passable in comparison to google. A couple of queries a week I am unsatisfied with and switch to google or a chatbot these days. I hope this new index is also comparable in quality and one step closer to a bit more competition in this market. Sadly, I could not find an official blog post or anything on the qwant news site..
mynti··on AI Twerk Generator
I clicked because i was very curious what this actually is, but it really is just what it says. The problem is, there is no twerking, the girls are just swaying from left to right really slowly which is not really twerking, but that is beside the point. Why are things like this being built? How can someone look themselves in the eyes after this?
mynti··on SensorLM: Learning the Language of Wearable Sensors
this seems very interesting: they got a big sensor dataset and generated some text from that. I guess this involves things like maximum values, mean values, maybe simple trends and things like was the person walking or biking etc. It would be interesting to see if the model identifies things that were not so easily provided in the training data. Otherwise this is just teaching the model to sort of calculate the mean from sensor data instead of using tools to do this
mynti··on Uber is testing a women-only driver service
this exact same thing happened in the show bojack horseman, made me chuckle
mynti··on A minimal ASCII art editor, place characters like pixels in a grid
i think this is quite neat! one feature would improve it for me quite a bit: if i just have a text document open i can easily click and input a character, so it is not much better there. but you could make it so that i can select a character and click and drag across the canvas, basically drawing the characters onto the canvas. that would be more fun imho
mynti··on IMO 2025 LLM results are in
> Grok-4 significantly underperformed compared to expectations. Many of its initial responses were extremely short, often consisting only of a final answer without explanation.

this is very weird. given how verbose most models usually are, there must have been something wrong in the system prompt.

also: grok used 89996 input tokens compared to 591624 for o3 high. What kind of tokenizer are they using that compresses the input so much? I suppose all inputs are actually the same, since the math problem + instructions are the same. only difference is the tokenizer or the system prompt. but i suppose it would not make up the difference. is o3 using 500k more tokens for their system prompt?

mynti··on Converting YouTube ad revenue into trees [video]
great energy, love the guy
mynti··on Grok is making AI companions, including a goth anime girl
someone who has never lived in a walkable city..
mynti··on Don't Eat Honey
i do not quite understand this point of view. even if bees suffer tremendously in the industrial complex, then the logical solution would not be to stop eating honey but to source honey from an ethical producer. there are lots of small beekeepers everywhere who treat their bees well and produce high quality honey. they get their hive in a often quiet area near a field or forest and live there mostly untouched until some of the honey is harvested. they also make sure to leave enough for the hive to survive, otherwise you would need to buy again.

i feel like this is the same for other animal products. for a lot of them there is an ethical way of sourcing them, which costs more and you would need to reduce your consumption of it. but this extreme view on veganism and animal products confuses me.

mynti··on Million Times Million
this is so funny because i always envied english for it being so clear: million, billion, trillion. in german we have these "awkward" names in between: million, milliarden, billionen, billiarden. but now hearing about this long scale it actually does make quite a lot more sense when thinking about it in multiples of millions
← PreviousPage 2 of 3Next →