Hacker Newsnew | past | comments | ask | show | jobs | submit | improbableinf's commentslogin

Opus 4.7 is non-usable for the tasks I have — but it’s considered an LLM.

And no one is stopping anyone from tweaking few parameters in this repo to go above 10M parameters.


What tasks is it non-usable for?


That’s exactly my thoughts. Code and documentation are one of the primary types of „content” by/for engineers. Kind of goes against the main topic of the article.


So you are not the part of a lucky 10,000 today…


According to downtector.com - both AWS and GCP are down as well. Interesting


Don't visit this address.


What a time to be alive!


When I read the comments I can’t really tell if it’s all a joke or not.


It's not a joke in the sense that the LLMs are just writing what they've been promoted to write...


That’s what someone who hasn’t disposed all of the coins would say.


“You can count calories and sugar content of the meal using just a camera”

“You can have a full self driving car with just a few cameras”

In a way both things are very much similar and the real accuracy is more of a fiction than reality.


Self driving car is far more plausible, we drive cars with visual input alone. But a camera can't chemically analyse the insides of your dish.


That’s exactly the point. We drive cars with visual input alone and have so many crashes, injuries and deaths every day.


An average person experiences very little car-related crashes, injuries and deaths every day.


> we drive cars with visual input alone.

That's unlikely. Try driving in a snowstorm, where visual inputs become effectively useless, and you quickly realize how much the motion inputs are factored in as well.


Ok, but motion inputs are even easier to record.


With the right sensors. Difficult to determine with visual input alone.

But with the right sensors you can figure out the chemical composition of food too, so...


They are very different problems. You can attach every kind of sensor to the phone and get HD images, lidar scan, 3D model, heat signature and whatever other signal you want, but you still won't be able to figure out calories in a piece of food without breaking it apart at a molecular level. There's just too much hidden information. What sensor will tell you how many tablespoons of oil the chef used to fry onions for the sauce? Heck the chef himself probably won't be able to tell you.


Hyperspectral camera?


I legitimately drove full month with near zero interventions in FSD. The second statement is achievable in my opinion. Most people have not experienced hw4 with latest releases.


FSD with just cameras for input breaks down in times of low visibility (fog, smoke, heavy rain). That’s a straightforward fact


That's an interesting comment, but I have driven in heavy rain, and it worked just fine. Can you share how you arrived at your conclusion? I mean there is a limit to where the obstruction is so intense that even human cannot drive I get that but its actually performing fine for me in heavy rain that is still drivable for a human.


How many people in my city or neighborhood must install the app (and use it, watch notifications) for it to serve its purpose?

It of course depends on the disaster, but in case of war this app can be used to find survivors for purposes other than saving them.

Great idea though.


Every person being rescued must have the app. Only one person rescuing needs to have the app.

It'll be easy to find a collapsed building site. The person rescuing just needs to be close to the site with the app. Otherwise, the signal might not be noticed.

+1 regarding finding survivors for malicious purposes. I tried my best to lower the surface area of that happening. It's still possible in edge cases, like opposing soldiers finding the people transmitting SOS, but they need to be extremely close to do that.


Does it work as well for remote workers?


Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact

Search: