1
214
Time to stop using Chrome (arstechnica.com)
submitted 1 year ago by Owl@hexbear.net to c/technology@hexbear.net

Google is now rolling out a system where Chrome directly tracks your activity and shares its summary with advertisers.

Also Firefox is faster as of like two months ago.

It takes five minutes to switch browsers, and the difference is so little that you'll often forget you did it.

2
5

title, basically. a comrade had shared his acc with me for a while (if you read this pm me! and thanks!!) without a hitch, until i managed to botch a cleanup of my PC because i was careless. I hadnt realized this until i deleted my hexbear acc with the og pm with the password and name lmao, so now im truly screwed. so yeah if anyone can help, id appreciate!!

3
15
4
17
5
162

Artificial Intelligence has enshittified the F-35, America’s long-embattled 5th generation stealth fighter jet. Produced by Lockheed Martin and coming in three different flavors, the F-35 relies on an AI system called the Autonomic Logistics Information System (ALIS) to maintain the aircraft. The pitch was that ALIS would keep the F-35 in flying shape cheap and easy.

This is an AI system that runs on a server that requires a room the “size of a shipping container” to function. It hasn’t gone well. According to the DOTE report, ALIS keeps telling support staff that things are wrong with individual F-35s when everything is fine. “Efforts to tackle the high false alarm rates have so far not yielded major progress towards meeting threshold requirements,” the report said.

The DoD has been attempting to use software filters to screen ALIS’s nonsense, but it doesn’t always work. And every time there’s a new piece of hardware or a software update, everything breaks again.

The F-35 has sucked for a long time. It’s got so many issues that it took the Pentagon 382 pages to elaborate on all of them. ALIS is just one of them. The Marines lost an F-35 in South Carolina last year. In 2021, one of the jets shot itself and caused $2.5 million in damage. Despite these and myriad other issues and crashes, the Pentagon is set to spend trillions on this thing and make millions more selling them to its allies.

6
57
submitted 1 day ago* (last edited 1 day ago) by hello_hello@hexbear.net to c/technology@hexbear.net

Did people really watch movies/shows on DVDs that forced them to watch ads before even starting? Like you go to the store and pay for a movie disc and when you go home you have to sit through like 10 minutes of ads. Did people really have to watch ads before they could even watch the movie they paid for a copy of?!

𝙏𝙝𝙞𝙨 𝘿𝙞𝙨𝙣𝙚𝙮 𝘿𝙑𝘿 𝙞𝙨 𝙚𝙣𝙝𝙖𝙣𝙘𝙚𝙙 𝙬𝙞𝙩𝙝 𝘿𝙞𝙨𝙣𝙚𝙮’𝙨 𝙁𝙖𝙨𝙩𝙋𝙡𝙖𝙮. 𝙔𝙤𝙪𝙧 𝙢𝙤𝙫𝙞𝙚 𝙖𝙣𝙙 𝙖 𝙨𝙚𝙡𝙚𝙘𝙩𝙞𝙤𝙣 𝙤𝙛 𝙗𝙤𝙣𝙪𝙨 𝙛𝙚𝙖𝙩𝙪𝙧𝙚𝙨 𝙬𝙞𝙡𝙡 𝙗𝙚𝙜𝙞𝙣 𝙖𝙪𝙩𝙤𝙢𝙖𝙩𝙞𝙘𝙖𝙡𝙡𝙮. 𝙏𝙤 𝙗𝙮𝙥𝙖𝙨𝙨 𝙁𝙖𝙨𝙩 𝙋𝙡𝙖𝙮, 𝙨𝙚𝙡𝙚𝙘𝙩 𝙩𝙝𝙚 𝙈𝙖𝙞𝙣 𝙈𝙚𝙣𝙪 𝙗𝙪𝙩𝙩𝙤𝙣 𝙖𝙩 𝙖𝙣𝙮 𝙩𝙞𝙢𝙚. 𝙁𝙖𝙨𝙩 𝙋𝙡𝙖𝙮 𝙬𝙞𝙡𝙡 𝙗𝙚𝙜𝙞𝙣 𝙞𝙣 𝙖 𝙢𝙤𝙢𝙚𝙣𝙩…

Even on VHS there were ads (you could fast-forward through them though), and Blu-Ray also has ads despite being a "more modern" standard (it's not it's just HD-DVD with a different branding). Also you can't even use the disc without paying for a special disc reader that reads that shit for you (tbf a lot of devices came with a disc reader, but it still persisted despite the fact that USB storage was far cheaper and more efficient). You'd also have to navigate the terribly slow menus just to get to the part you were at.

Also if you buy a DVD/Blu-Ray whatever the fuck they call it nowadays in one part of the world and you travel to another, say you have family that lives in one country and you live in another, you can't play that disc because it's "region-locked."

Ok maybe it's region locked because different countries probably use different displays/standards or whatnot. NO! It's region locked for NO MATERIAL REASON besides "ensuring copyright distribution of the holder". This is even more mind-boggling for "blu-ray" the supposedly new format.

Also most Blu-Rays don't even come with all the goodies that normal DVDs had like behind the scenes/deleted scenes etc, so it's not like Blu-Rays have any other advantage besides being incompatible with your dvd player. "Just buy a PS3" yes I will buy the SONY product to play movies on a disc also created by SONY.

How is it considered physical media when the devices to play it are not being sold anymore? I'm sure there are a lot of Sony walkmans being sold nowadays. I can totally pick up a VHS player right now at the store and enjoy my treasure trove of vhs tapes that haven't already withered to dust.

People older than me (I was born after Al Gore lost the election) are having nostalgia for the "age of physical media" when really it was an age of physical bullshit compared to streaming bullshit. It's always capitalism, capitalism will burn down all art if it means that someone didn't get to skip paying for it. Here's what I say, just pay a couple a dollars a month for a VPN with port-forwarding and just torrent all your media. Your torrented file has done more for media preservation and archival then any DVD bullshit ever did. The only use for physical media is to digitize it and share it.

The bootlegged Cinderella movie sold in the Global South has done more for media preservation than Disney ever has. A seedbox in Russia is more of a art library compared to any video store.

Don't get me started on video games. Where every generation of devices there's a new standard and new way to do things. Nothing says media preservation like buying a disc from a store and then waiting an hour for your device to download updates online.

7
24
8
36
9
63

I have the sudden urge to try my hand at 3D art.

I'll make something 3D and post it here hehe.

OOOH I could make us some crusty emotes!

Is blender hard?

10
42
11
15
submitted 3 days ago by miz@hexbear.net to c/technology@hexbear.net

please share what you're using to manage your photos.

digiKam? PhotoPrism? photostructure? resourcespace?

12
35
13
43

I don't even like micro-blogging in general - a "twitter replacement" is the last thing the world needs but I signed up for bsky out of curiosity and feel like ranting about how god-awful the UX truly is.


The "Discover" feed that you start out with is absolutely useless. The site asks you for a list of interests when you sign up, which seems to determine what shows up in Discover, but there is no way to change your interests afterwards. also Discover shows lib-ass posts even if you didn't select politics or news as interests, so fuck that. The good thing is you can remove the discover feed entirely.

"Feeds" is in interesting concept but is confusing and badly implemented. Feeds can work on different criteria - they can include posts from certain accounts and/or posts that contain certain keywords / hashtags. Or it can be a more complex user-written algorithm. But there is no way of seeing what criteria a particular feed operates on, it's a total black box.

Feeds are not something a user can easily create - you have a write code and run a server. This is obviously a major problem and as long as this is the case, feeds will be geared towards the interests of tech/nerd types.

There are three different things you can do to a feed - like it (pointless), save it which adds it to "My Feeds", or pin it which adds it to the horizontal scroll of feeds as well as the vertical list in the sidebar.

The horizontal feed scroll is stupid and works like shit on desktop - dragging with a mouse is clunky as you always end up accidentally highlighting text when you get to either end of the scroll. For the love of god, just give me a dropdown.

Clicking the + icon in the feed discovery page actually saves and pins that feed. To save a feed without pinning it, you have to go to the feed's page, click on the ••• dropdown and click "Save to My Feeds" which also uses the + icon. So depending on the context, + can either mean "save" or "save and pin" apparently.

Btw why in the fuck do "feeds" use the hashtag icon??

There are also "lists" which is just a list of users, and "starter packs" (stupid term) which is a list of users and feeds - why are these two different things?


Worst of all, there are no communities on bluesky. Shouldn't there be like, groups that you can join and see all posts from others who have joined that group? Nope, much like twitter the experience is of course hyper-individualized. You can create your own personalized feeds, lists, or starter packs, or view other people's personalized feeds, lists, or starter packs, but you can't really belong somewhere. There is no community, only curation.

14
141
15
42
16
32

My phone is dying and i also really want to increase my privacy and fully decouple myself from the google ecosystem so i was thinking of moving to a pixel and using grapheneOS. Now is the time before tech prices shoot way up from tariffs. What model of pixel should i get that isnt too expensive but isnt likely to lose grapheneOS support anytime soon? I was thinking like a 7 maybe

17
42
18
25

...or at least the kind of model that isn't fueled by burning a small forest for every query? I am wanting to play an old video game, but I'm still only just learning the language and could use an aid. I really, really want to avoid any of this incredibly wasteful AI stuff.

19
14

...or at least the kind of model that isn't fueled by burning a small forest for every query? I am wanting to play an old video game, but I'm still only just learning the language and could use an aid. I really, really want to avoid any of this incredibly wasteful AI stuff.

20
14
submitted 1 week ago* (last edited 1 week ago) by ahriboy@lemmygrad.ml to c/technology@hexbear.net

It wasn't a health concern. The only concerns for 5G were long-term practicality, energy and service costs and shorter coverage.

5G was supposed to be made for robots, IoT and wireless home internet.

I haven't tried 5G, but it seems that the expected speeds would be the same as 4G in early days.

21
61

It feels like the US isn’t releasing what it has. I don’t think they’re behind, maybe just holding back?

i-cant

22
12

Test-time training (TTT) significantly enhances language models' abstract reasoning, improving accuracy up to 6x on the Abstraction and Reasoning Corpus (ARC). Key factors for successful TTT include initial fine-tuning, auxiliary tasks, and per-instance training. Applying TTT to an 8B-parameter model boosts accuracy to 53% on ARC's public validation set, nearly 25% better than previous public, neural approaches. Ensemble with recent program generation methods achieves 61.9% accuracy, matching average human scores. This suggests that, in addition to explicit symbolic search, test-time training on few-shot examples significantly improves abstract reasoning in neural language models.

23
41

AI, as it currently exists, is terrible, devoid of much real value or use, and probably overall degrading our society (and definitely our climate). This is unlikely to change as the technology requires massive amounts of capital to train and run the models. Massive data centers that are an endless pit of energy, ghouls that want to use the technology for nefarious purposes such as extracting more wealth from you and I, to the extreme of facial recognition of "terrorist" suspects for drone strikes.

There is massive amounts of capital in this area, and other than the energy and equipment for running the models, another big cost for firms that develop LLMs is the need for training data that requires hundreds to thousands of people to generate manually. This is where you can potentially earn some decent remote money. There are caveats, and I by no means recommend this to anyone as a sole source of income. Work availability fluctuates week by week, the biggest company in this space is very evil, and you might end up spending a lot of time for very little benefit.

Your actual ability to earn depends mostly on your country of residence your level of education. In the USA, with a M.S. in Chemistry (they have never confirmed any of my education btw), I have been able to earn at most $60/hr, but more often $25-35/hr. The pay changes constantly because every project pays differently, and you will change projects very often, usually every week or two. I have also gone through a 2 month dry spell of having no/little work to do, which is why this is not something I would ever recommend as a source of income you rely on.

What is AI training and why should I care?

Specifically, AI training encompasses any work that involves creation or annotation of data that is fed back into an LLM model to improve it's capabilities. The data can be images, audio, video, and most commonly text. The specific task at hand can vary widely, but the most common involves rating various dimensions of the LLM's response and then improving the response.

How exactly is this used by a company? As a simple example, imagine that you are creating a service that uses an LLM to create recipes from a list of ingredients. First, you might download every cookbook and recipe that you can find by crawling the web, and feed that into the model. Unfortunately, the quality of the recipes might not be very well standardized, some transcription errors may have occurred, or you might have other specifications that you would like your model to follow. So you send those recipes, one-by-one, to thousands of people who will fix any errors, add additional context, and make sure that the recipes fit the specifications. Now this data set can be used to fine-tune the model and hopefully improve it by some measure.

Companies I have worked for

Scale Labs (Remotasks, Outlier)

Scale Labs is by far the largest company in this space, and is, at least up until recently, the de facto monopoly. It was founded by Alexandr Wang in 2016, after he dropped out from MIT at the age of 19. The company grew by setting up large centers (aka digital sweatshops) in Sub-Saharan Africa and Southeast Asia, paying the lowest possible wage for workers to annotate images from autonomous vehicles. Imagine sitting in a sweltering computer center, getting paid well below $1 an hour, to circle cones, pedestrians, vehicles, bikes, over and over again, with no job security - that is how this little psychotic dipshit become a billionaire at the age of 24. There are numerous articles out there about how bad this company is, and of course they are also the official AI partner of the DoD.

Scale runs two tasking platforms - Remotasks, and Outlier, the latter of which myself and @corgiwithalaptop have used. They seem focused on recruiting people with specific expertise, for a list of all fields you can look at their site (outlier.ai). Although I was hired as a "Chemistry Expert", most of my projects haven't involved Chemistry at all.

There are some major problems with this platform:

  • You will have no tasks (empty queue, or EQ) for extended periods of time (sometimes weeks to months)
  • "Training" is usually unpaid and extreme bullshit
    • You need to do new training for each new project. The training documents are often vague or contradictory. You have to do graded quizzes and graded assessment tasks that you will fail, and never get feedback on why you failed. This is a very common experience, I once had a single week where I failed 4 assessments in a row, meaning I did 4 unpaid trainings for no reward.
  • The platform itself is full of bugs and technical issues
    • Once I was assigned to a project, but a bug prevented tasks from actually being allocated to me. The project's managers (queue managers, or QMs) told me that it is a bug that I would have to ask support about. By the time support got back to me over a week later, the project had wrapped up. I got assigned to a new project, and faced the exact same bug, which took another week to get fixed. This entire time, I am unable to do any work on the platform or make any money.
    • If there is a technical issue that prevents you from completing a task, you will not get paid at all for the time you had already invested in the task.
  • Every task gets reviewed and you are given a rating out of 5. Reviewers are just normal taskers that get promoted to reviewer willy-nilly (I have been a reviewer on many projects), and some reviewers are just god awful. They will straight up not understand your prompt, or not be familiar with some concept in the task, and give you a 1/5 for no reason. I have even been on projects where people were using an LLM to automatically complete the reviews, giving erroneous bad scores. If you get enough low scores, you can be removed from the project.
  • Support is non-existent and useless. It takes days to weeks to get any response, and sometimes the response is just a canned response where you can tell they didn't actually read your ticket.
  • They can remove you from the platform at any time for any reason, and anecdotally I have heard about this happening to people who did nothing wrong. There is no recourse if this happens to you, you can reach out to support but like I said, this is useless

Despite these problems, I am currently making money on the platform and will continue to do so as long as I can. For all of these reasons, I can not recommend outlier unless you have infinite spare time and really want a remote way to make money. If you have an expertise, especially coding/math, I think you might have a better chance of getting in and making good money. Recently, they have been promising to make changes to address these issues, but scale labs is ultimately an authoritarian employer that has no incentive to make their workers lives better, and only has incentive to increase the efficiency of labor extraction, so take that with a grain of salt. If you need income, I would not recommend wasting to much time on it or expecting anything out of it, but it might be worth a try.

Stellar AI

I have only been on this platform for about a week, and am only on one project so far, so I don't have much to say other then the following ways it is better than Scale's platforms:

  • Training was much better. Instead of just reading a google doc, they walk you through exactly how to do a task within the tasking interface. It was also paid
  • If there is a technical issue that prevents you from completing a task, they will still pay you for your time My current project involves guiding the AI through navigating websites to get information to answer a prompt, for example, "what is the cheapest airline to fly to Hawaii," it is not all that difficult and the pay is $25/hr.

Conclusions

If you are good at writing-based work, and want remote ways to make money, and have plenty of spare time, it might be worth trying out these platforms. The worst case scenario is that you spend some time on it and ultimately reap no benefit, but there is a chance you get in and can successfully make decent money doing work that really isn't that difficult. The majority of the stress I have had is due to the Outlier platform itself.

If you have special expertise (any advanced degree, multiple languages, especially if you have coding experience) you might have better chances. Pay rates vary by country, so if you aren't based in the USA I am not sure what the earnings potential is.

@corgiwithalaptop also works for Outlier and they might be able to offer a different perspective. If you do decide you want to apply for Outlier, reach out to me or them and we can give a referral code - if you use my code and get the job and complete 10 hrs of work I would get $200, which I pledge to give $100 through the mutual aid comm here and the other $100 to my local mutual aid.

24
99
25
16

Unlike traditional language models that only learn from textual data, ESM3 learns from discrete tokens representing the sequence, three-dimensional structure, and biological function of proteins. The model views proteins as existing in an organized space where each protein is adjacent to every other protein that differs by a single mutation event.

They used it to "evolve" a novel protein that acts similarly to others found in nature, while being structurally unique.

view more: next ›

technology

23313 readers
279 users here now

On the road to fully automated luxury gay space communism.

Spreading Linux propaganda since 2020

Rules:

founded 4 years ago
MODERATORS