For contrast, I used a Surface Book throughout college and within weeks of the warranty expiring I ran into serious issues with the battery, then the charging port, display backlight, fan. I loved it to death so I kept it on life support and changed my usage patterns until I gave up on it. And yes, my next device was a used Thinkpad, and I was able to fix most issues I ran into. But I'd

I am NOT a fan of the measures Apple takes to monopolize the maintenance and repair of their devices.

In 2023, I was playing with transformers, RNNs and I had an understanding how it worked from top to bottom (e.g. made my own keras, could whiteboard small nets) and I can throw things together in keras or tf pretty quick

I got a job and never touched that again. Data and compute notwithstanding, how hard would it be to make a pet project foundation model using the latest techniques? I’ve heard about MoE, things like that and I figure we’re not just throwing a bunch of layers and dropout in Keras anymore.

Comments URL: https://news.ycombinator.com/item?id=46236222

Points: 4

# Comments: 3

> Nvidia has some managers who are telling their people to use less AI

What's happening now is already pretty incredible given the understanding that we're basically still at the 'chat bot' stage for most people. The idea of agency is still very, very recent, but it's understandable that most people (particularly non SDs) are not impressed.

It's easy to look at the present and be cynical. If it is only able to solve your problem 95% of the time, you still can't trust it. I think the bets are really about how far we are from 99%, even for random stuff. The fact that a chatbot that was never explicitly trained to, just by predicting next probable tokens is wild. The pace of improvement in the past 5 years has been dizzying.

I'm not out here trying to put a dollar amount on it. But certainly, there is going to be a lot of money to be made. Of course it's a front for money and power. But like... isn't that the point of a corporation?

Correction. $0 in value. Skins do not exist and are worth exactly $0. If you spend money on skins, they are worth… $0. It’s all a large scale grift money incinerator where the only winner is Valve.

+ whatever pleasure you derive from it, ig. I can understand loot box addiction, but paying $20,000 for valve character dress up? Not even like a Peter Griffin player model or something, but a slightly different looking knife? Madness

Persp: tf2 enjoyer

Findings summarized:

They added a chat window with a half-rate model and access to OneDrive and sharepoint. Obviously useful integrations like setting up meetings based on a thread or creating tasks are not present.

I don’t care about email summaries or drafts. I talk with developers and our emails are precise.

Github copilot: The shared naming is confusing. There’s not even a shared UI language between the two. I use it because I don’t have Claude code at work. They’ve added some little things like pulling errors from the terminal but it’s mostly just a chat window. Meh.

Also if I edit the file outside the editor (maybe stash some changes with git) before approving edits things get broken and the LLM will start adding in old changes and it becomes part of the context and it’s kind of a mess.

Copilot (browser): -separate ‘work’ and ‘web’ mode for some reason. As far as I can tell it can see my web page in ‘web’ mode but I lose my OneDrive and sharepoint junk. Minimal utility. ChatGPT is better.

Outlook/teams: It’s totally inconsistent and missing integrations. Long email thread? Let’s use Copilot to avoid having to draft a meeting invite with everyone on the thread! There’s a context menu option to do this!

Oh, ok. They removed that in the past week.

Oh ok. It has no context of what I have open. I have to describe the email I’m currently in to their sidebar. This is already not worth my time.

Hmm. It made zero attempt to schedule a time where everyone is available.

This goes on, and I wanted to try because scheduling with our big wigs is a pain. No dice.

Post meeting. We’ve got takeaways. Convert these to MS Todo tasks? No dice! No integration. Want to find teams messages or schedule from teams? Also no.

Excel: neat for making formulas. But sometimes I want to do LLM classification or structured response on a tabular level. That’s the next thing. Also includes a useless chat window.

Powerpoint: also useless. Strangely, I can’t use another powerpoint as a reference for a powerpoint. I want to create clipart. Make a theme. They have ‘make a slide with copilot’ that I actually like, if I could use what it generates as a theme and apply it for all slides.

Generating a slide deck based on a bunch of input is actually pretty cool - but the generic ‘synergy business’ lingo and clip art it uses is so cringe.

> People want to log in to an account, tell the thing to do something, and the system figures out the rest.

At a glance, this seems to be a practical approach to building up a personalized prompting stack based on the things I commonly do.

I’m excited about it.

I definitely see niches for both. Even if you've got some experience an Arduino uno or mega is just an atMEGA with good software support and IO headers.

We'd usually use an RPI and Arduino - connect our 'out of the box' modules to the pi, pi to arduino via uart serial, and wire arduino to the meat and potatoes. The RPI's IO was generally not as good in terms of latency but also if the wrong wire gets crossed suddenly we'd have a dead Pi but the Arduino would shrug it off.