kadushka

454 karma · joined May 9, 2017

kadushka··on Maine’s ‘Lobster Lady’ who fished for nearly a century dies aged 105

Can you imagine writing code for 100 years?

kadushka··on Vitamin D and Omega-3 have a larger effect on depression than antidepressants

the larger the trial size, the smaller the outcome

I find this a bit surprising. Could there be something else affecting the accuracy of larger trials? Perhaps they are not as careful, or cutting corners somewhere?

kadushka··on Will AIs take all our jobs and end human history, or not? (2023)

This will break down when >30% of people are unemployed

kadushka··on An Ice-Covered Russian Ghost Town

https://nord-ursus.livejournal.com/171473.html

kadushka··on The Illustrated Transformer

Maybe so, but this particular blog post was the first and is still the best explanation of how transformers work.

kadushka··on GPT-5.2

What kind of improvements do you expect when going from 5 straight to 6?

kadushka··on Cloudflare Global Network experiencing issues

I honestly wanted to understand your position, but after such a reaction, I'm not going to engage in any discussions with you.

kadushka··on Cloudflare Global Network experiencing issues

Your position is similar to saying that medical drugs have been a net negative on society, because some drugs have been used and abused to collective detriment (and other negative effects, such as doctors prescribing pills instead of suggesting lifestyle changes). Does it mean that we would be better off without any medical drugs?

kadushka··on Cloudflare Global Network experiencing issues

Are you anti-AI in general, or are you unhappy about the current LLMs?

kadushka··on Cloudflare Global Network experiencing issues

Would you rather not have LLMs?

kadushka··on AGI fantasy is a blocker to actual engineering

Connections between LLM neurons also change during training.

kadushka··on 'Attention is all you need' coauthor says he's 'sick' of transformers

devil is in the details

kadushka··on OpenAI's hunger for computing power

Mainly because global video data corpus is > 100k larger than global text corpus, so you will need to train much larger models for much longer (than current LLMs).

kadushka··on The collapse of the econ PhD job market

Yes, this fact is well-known and has been widely discussed. The first two google search results provide the stats:

https://students.bowdoin.edu/bowdoin-review/features/death-b...

https://www.forbes.com/sites/paulweinstein/2023/08/28/admini...

kadushka··on The collapse of the econ PhD job market

I hope they're cutting their bloated administration departments before cutting classes and lecturers, but I'm not holding my breath.

kadushka··on Terence Tao: The role of small organizations in society has shrunk significantly

Was that the same Donald Trump who said he would end the war in Ukraine in 24 hours if elected?

kadushka··on Terence Tao: The role of small organizations in society has shrunk significantly

What subsidies is Stargate project getting from US government?

kadushka··on Trevor Milton's Nikola case dropped by SEC following Trump pardon

Woz did make billions though

kadushka··on DeepMind and OpenAI win gold at ICPC

it was not a revolutionary update but evolutionary at best compared to o3

It is a revolutionary update if compared to the previous major release (GPT-4 from March 2023).

kadushka··on CubeSats are fascinating learning tools for space

So it costs $85k to launch such cubesat. Too expensive for almost all of us. But if the cost comes down to say $5k, I'd probably be interested in this as a hobby project.

kadushka··on Australian startup joins race to build local ChatGPT

256 of the latest Nvidia Blackwell B200 GPUs

Did they forget to add "k" to that number? OpenAI plans to have one million GPUs by the EOY.

kadushka··on Why is Japan still investing in custom floating point accelerators?

it's "one really smart person works for a few weeks" hard

AMD should hire that one really smart person.

kadushka··on GHz spiking neuromorphic photonic chip with in-situ training

Maybe try simulating the algorithms in software before building hardware? People have been trying to get spiking networks to work for several decades now, with zero success. If it does not work in software, it's not going to work in hardware.

kadushka··on Cerebras launches Qwen3-235B, achieving 1.5k tokens per second

Could there be a better

Yes. Look up “block floating point”.

kadushka··on How to break the 'AI hype cycle'

Funny to see this on the front page right next to an announcement of a second model getting gold on IMO.

kadushka··on AI capex is so big that it's affecting economic statistics

Frontier models like o3 reason better than most humans. Definitely better than me. It would wipe the floor with me in a debate - on any topic, every single time.

kadushka··on Agents built from alloys

I always do this with o3, gemini 2.5, and opus 4 when brainstorming hard problems: copy each model’s response to the other two.

kadushka··on Ask HN: Having terrible time with paid versions of ChatGPT and Claude

Opus and o3 usually work ok for me, meaning they get the job done with some steering.

kadushka··on Master Foo and the Script Kiddie

I’m curious what was the intended lesson to crack that machine.

kadushka··on AI capex is so big that it's affecting economic statistics

I don’t know which llm you used - I just asked gpt-4.1 - it did a web search and provided the correct exchange rate. It took about 5 seconds.

← PreviousPage 2 of 12Next →