HNHacker News
TopNewBestAskShowJobs

KaoruAoiShiho

3,254 karma · joined October 1, 2011

Looking for investors & partners: kaoruaoishiho@gmail.com
submissionscomments
KaoruAoiShiho··on Hot take: GPT 4.5 is a nothing burger
That's obviously untrue... Yes it hasn't hit the usable threshold for human replacement for many tasks, the majority of tasks, but that doesn't mean it hasn't improved massively for many others. Your evals seem to just be not fine enough, probably you're just waiting for agi. I do evals on a ton of tasks and my API/chatgpt usage has shot up drastically. It's now irreplaceable tool for me.
KaoruAoiShiho··on AI is killing some companies, yet others are thriving – let's look at the data
can't get from LLMs for 6 months until the next training run?
KaoruAoiShiho··on Claude 3.7 Sonnet and Claude Code
The copying here probably goes to strawberry from o1 which is like at least 6 months but maybe copying efforts started even earlier.
KaoruAoiShiho··on Claude 3.7 Sonnet and Claude Code
They cited Cognition (Devin's maker) in this blog post which is kinda funny.
KaoruAoiShiho··on Kaneo – An open source project management platform
Broken after adding a project.
KaoruAoiShiho··on Some critical issues with the SWE-bench dataset
Me too, cursor+sonnet is also my go to, I just didn't really understand what you were getting at by pointing out this benchmark. I guess it is significant that Sonnet is the actual line by line coder here. It is the best at that, and it's better than DeepSeek+any other combination and better than Any other reasoner+Sonnet.
KaoruAoiShiho··on Some critical issues with the SWE-bench dataset
Sonnet is literally lower on the aider benchmark you just linked. It's only the top with Deepseek as architect, otherwise it's lower than many others.
KaoruAoiShiho··on What enabled us to create AI is the thing it has the power to erase
That's why it's so important for R1 to show the reasoning. The creative process has to be transparent and easily understood.
KaoruAoiShiho··on Revenge of the GPT Wrappers: Defensibility in a world of commoditized AI models
Idea guys are a dime in a dozen.
KaoruAoiShiho··on Revenge of the GPT Wrappers: Defensibility in a world of commoditized AI models
I predict this article to be embarrassingly wrong. The moat of models is compute, wrappers are just software engineering, one of the first things to be commoditized by AI in general.
KaoruAoiShiho··on GenAI Art Is the Least Imaginative Use of AI Imaginable
I don't disagree at all, the process of human storytelling is key to art. But at the same time I'm still going to use genAI 99% of the time when I just want a pretty picture, when I'm in the mood for doing it myself I will, but most of the time I probably won't be.
KaoruAoiShiho··on DeepSeek R1 Is Now Available on Azure AI Foundry and GitHub
How's the pricing as compared to official API or openrouter providers?
KaoruAoiShiho··on TikTok goes dark in the US
Are you sure about this? I've heard many many times that tiktok is uniquely good at discovery for new businesses.
KaoruAoiShiho··on Proof of location for online polls
Did you read the article? It's not about using IP addresses...
KaoruAoiShiho··on Proof of location for online polls
Fantastic, solves the issue of bots from foreign adversaries. Everyone complaining doesn't seem to get it, it doesn't need to solve all usecases, but solving this one usecase is great.

Conversely, can this be used to show that someone is NOT a chinese/russian bot? I've had enough with people accusing me lol.

KaoruAoiShiho··on Learning how to think with Meta Chain-of-Thought
I'm sure if you take an hour to recall you'll be able to come up with a process. Or ask a philosophy professor who specializes in reason.

But the easiest way I can think of ATM is to go through all the questions that AI currently fails on, and then have a human work through them and show the chain of thought a human would do, including the false starts, and describing the strategy pivots. Then generate your corpus based on that. However, that burns the problem-set so you'll have to constantly try to come up with new problems.

KaoruAoiShiho··on Learning how to think with Meta Chain-of-Thought
Just train it on meta reasoning, ie train it on people discovering ways to discover. It's not really a big problem, just generate the dataset and have at it.
KaoruAoiShiho··on OpenAI O3 breakthrough high score on ARC-AGI-PUB
o1 already fixed the red herrings...
KaoruAoiShiho··on Veo 2: Our video generation model
ByteDance has their own unlimited supply of videos...
KaoruAoiShiho··on Microsoft Recall still storing credit card, social security numbers
Ironically Recall is what made me go back to windows.
KaoruAoiShiho··on Trillium TPU Is GA
Is there not goin to be a v6p?
KaoruAoiShiho··on Gemini 2.0: our new AI model for the agentic era
It looks like long context degraded from 1.5 to 2.0 according to the 2.0 launch benchmarks.
KaoruAoiShiho··on AI Does NYT Connections
QwQ gets it wrong, Gemini gets it wrong. o1 gets it right, R1 gets a pretty good not originally intended set of 4... tempted to give it partial credit. 4o gets it wrong. Will update with claude once my usage limits are up lol.
KaoruAoiShiho··on AI Does NYT Connections
Took a bit of prompting and a few awful ones, but ultimately it's not bad.

https://chatgpt.com/share/67570ab1-b2c0-8006-b5d2-d3fa7132de...

Going to try to feed this into some other LLMs like qwq and see if they can solve them.

KaoruAoiShiho··on AI Does NYT Connections
Beyond being able to solve Connections, can a LLM generate (good/challenging/solvable) connections? Would be pretty cool to be able to generate a test set.
KaoruAoiShiho··on Next-level frosted glass with backdrop-filter
Wait is this better than the one in the OP?
KaoruAoiShiho··on Next-level frosted glass with backdrop-filter
How does this compare with this that was posted on HN a few weeks ago?

https://www.tyleo.com/html-glass.html

KaoruAoiShiho··on AI Stole the Joy of Programming
It probably varies by environment, all I know is that it works 98% perfectly for me.
KaoruAoiShiho··on Genie 2: A large-scale foundation world model
This is where the GPU limits on China really hurts, Chinese companies have been dropping great proof of concepts but because they have been so compute bottlenecked they can't ever really make something actually competitive or transformative.
KaoruAoiShiho··on AI Stole the Joy of Programming
That's because you don't find joy in programming, you find joy in game design. Actually, game design can be automated too just the interfaces are not as advanced as cursor. We'll get there soon.
← PreviousPage 4 of 34Next →