ROCm ships its installable client driver (ICD) loader and an OpenCL implementation bundled together. As of January 2022, ROCm 4.5.2 ships OpenCL 2.2
Shaders are computational visual [post-]processing - think pixel position based adjustments to rendering.
OpenCL and CUDA are computation frameworks where you can use the GPU for other processing than rendering. You can use it for more general computing.
nVidia has always been focusing on proprietary technology. Introduce a technology, and try to make it a closed market, where people are forced to buy and use nVidia for it. AMD has always been supporting and developing open standards as a counterplay to that.
JWTs are standard authentication tools - who’s the security concern for? ByteDance? Or are you saying the JWTs are from the local machine?
Yes, I read that as local project JWTs are being transmitted to their servers. As a concern, and not labeled as used for authentication, IMO it's clearly implied that they observed JWT tokens and auth data unrelated to any telemetry auth (if they even have any).
JWT tokens and authentication data observed in multiple communication channels, presenting potential credential exposure concerns
Arguably, the openness is in that the EU OS can switch from one to another at some point if it becomes necessary.
Supporting multiple alternatives within the same platform and OS is costly. Not only the integration, but also user training and troubleshooting, specifically about the many, big and small subtle differences. Focusing on one, for now anyway, makes sense.
The math guessing game (lol), the bullshitting of "thinking out loud", being able to identify hidden (trained) biases, looking ahead when producing text, following multi-step reasoning, analyzing jailbreak prompts, analysis of antihallucination training and hallucinations
At the same time, we recognize the limitations of our current approach. Even on short, simple prompts, our method only captures a fraction of the total computation performed by Claude, and the mechanisms we do see may have some artifacts based on our tools which don't reflect what is going on in the underlying model. It currently takes a few hours of human effort to understand the circuits we see, even on prompts with only tens of words.
I would separate concerns. For the scraping, I would dump data as json onto disk. I would consider the folder structure I put them into, whether as individual files, or a JSON document per line in bigger files for grouping. If the website has good URL structure, the path could be useful for speaking author and or id identifiers in folders or files.
Storing json as text is simple. Depending on the amount, storing plain text is wasteful, and simple text compression could significantly reduce storage size. For text-only stories it's unlikely to become significant though, and not compressing makes the scraping process, and potentially validating completeness of scraped data simpler.
I would then keep this data separate from any modifications or prototyping I would do regarding modification or extension of data and presentation/interfacing.
JSON is a much simpler (and consequently safer) format. It's also more universally supported.
YAML (or TOML) is decent for a manually read and written configuration. But for a scraper output for storage and follow-up workflows being through code parsing anyway, I would go for JSON.
Unless you continuously change you IP I don't see how locking DNS resolution behind a signup would solve it. You only need to resolve once, and then you know the mapping of domain to IP and can use it elsewhere. That mapping doesn't change often for hosted services.
Any wall you build up will also apply to regular users you want to reach.
Infrastructure configuration that is automatically applied to the cloud infrastructure. Like starting and stopping new instances and services, changing connections between them, etc. (I assume anyway.)
If it's the one that makes the most sense to you then do it.
Even if the job market is at a low, I doubt other industries fare much better. If you'll be a good dev you'll likely be in good demand. (Even then job hunting may be a hassle, but I doubt that's different elsewhere.)
Go for it! It's not like you're stuck on that road forever anyway.
Not much more than the stock images we had before. When they're added to articles, most of the time, they added nothing but noise and wasted layout space the same way. It's just that the image is generated, sometimes [even more] silly now.
I'm usually not using bash locally, and remotely don't change the prompt, but Starship works in bash too.
I use Nushell with Starship (cross platform prompt) in Windows Terminal.
~
nu ❯ took 52ms
Path above prompt, prompt with shell name and a character, and on the right side how long the previous command took. The Character changes color from green to red when the last command exited with a non-0/-success exit code.
In a git repo folder it shows git info too - the branch symbol won't show here because here is not a nerd font with symbols; I'll add a screenshot:
C:\dev\dotnet\meercat-monitor on main [?]
nu ❯ took 1ms
Starship can show a bunch of status/state information for various tools, package managers, docker, etc.
I wouldn't show my PROMPT_COMMAND, but it's a nu closure so not really comparable to bash. But as I said, Starship works with Bash too.
and include expensive endpoints like git blame, every page of every git log, and every commit in your repository. They do so using random User-Agents from tens of thousands of IP addresses, each one making no more than one HTTP request, trying to blend in with user traffic.
That's insane. They also mention crawling happening every 6 hours instead of only once. And the vast majority of traffic coming from a few AI companies.
It's a shame. The US won't regulate - and certainly not under the current administration. China is unlikely to.
So what can be done? Is this how the internet splits into authorized and not? Or into largely blocked areas? Maybe responses could include errors that humans could identify and ignore but LLMS would not to poison them?
When you think about the economic and environmental cost of this it's insane. I knew AI is expensive to train and run. But now I have to consider where they leech from for training and live queries too.
ROCm is an implementation/superset of OpenCL.
Shaders are computational visual [post-]processing - think pixel position based adjustments to rendering.
OpenCL and CUDA are computation frameworks where you can use the GPU for other processing than rendering. You can use it for more general computing.
nVidia has always been focusing on proprietary technology. Introduce a technology, and try to make it a closed market, where people are forced to buy and use nVidia for it. AMD has always been supporting and developing open standards as a counterplay to that.