TrueDuality

I'm running P41s in one of my test boxes. These don't have support for BF16 but they do support F16 and F32 and those are accelerated to a certain degree, they're lacking kernels that are as optimized but its not terribly hard to adapt other ones for the purposes.

You don't get great out-of-the-box performance but it only took me three work days or so with no experience writing these to adapt, test, and validate a kernel using the acceleration hardware that was available (no prior experience writing these kernels).

They're not as powerful as others but still significantly better than running on a CPU alone and I'd bet my kernel is missing more advanced optimizations.

My issue with these was the power cable and fans. The author touches on the fans and I did try a 3D printed shroud and some of the higher pressure fans but I could only run the cards in short stints. I ended up making an enclosure that went straight out of the case using two high pressure SAN array fans I harvested from the IT graveyard per card and making a hole with an angle grinder.

The power cable is NOT STANDARD on these. I had to find a weird specific cable to adapt the standard 8-pin GPU connector and each card takes two of these bad boys.

As long as Apple has a business presence in the UK, they are subject to the laws the UK imposes on them even if they're vastly overreaching and impose on other government's citizens. Not supporting cloud services wouldn't be sufficient to avoid the compliance requirement, they would have to formerly stop doing business in the UK.

Looking at the market size that might be a decision that Apple is willing to make as it would most likely be a temporary stick. The government can spin it anyway they want, but Apple devices do not work basically at all without the deep integration of their services. A geoblock would effectively mean UK citizens would be left with unusable devices and I can't see the resulting outrage being directed exclusively at Apple.

It'll be interesting to see how this plays out for sure.

Very cool! This definitely has the potential to make eavesdropping on strangers significantly more accessible. I'm a tad worried about this kind of proliferation but this kind of thing is probably inevitable and getting something out there at least lets people to start discussing it openly with a concrete reference.

I really like interacting with models via voice while typing through other apps but its infeasible in loud environments, and feels bad and kind of invasive in places like airports. Looking forward to trying this out more aggressively.

This is a bunch of slop. The model seem interesting but the article has very very little content, a lot of fluff, and some pushes of an unrelated tool.

One gotcha that annoys me when do this... The file still needs to end in .py which I generally don't like for the bins and utilities in my path. Everyone using these shebangs either doesn't hit this or never mentions it. Entirely a stylistic personal preference but still makes me a sad lady.

This is another great call-out and semi-common. I can definitely get blinded by my security focus but shady business tactics drive a lot of these similar domain purchases for exactly the reason you described.

Yeah this is a good call-out. If the site is being used for drive-by or targeted malware there are other checks that may be happening alongside the redirect such as user agent, country of origin (like you mentioned), plugins installed, OS, or even time of day.

If they detect something that matches what they want, they may throw some intermediate 301's to pages that attempt to infect the user with something still ultimately redirecting to the "normal" page.

As others have mentioned this is likely one of a couple of scenarios, roughly ordered by my guess on likelihood:

- Attempting to use your legitimate content and services to improve the SEO rank of other domains (even unrelated ones). This can usually be checked by looking for a sitemap.xml, there will be pages not redirected to your site that contain pages of links.

- Closely following the above, the pages may not be links to other sites but might be hosting phishing pages for other services unrelated to yours. The redirect here acts as a bluff for casual inspection of the domain. You won't see page entries in a sitemap.xml file for these ones.

- Attempting to "age" a domain. Not many talk about this option, but new domains are a red flag to a lot of automated security processes. When purchasing a domain and giving it a history associated with a legitimate service they make the domain look less suspicious for future malicious use.

- Preparation for a targeted campaign. This is pretty unlikely, you need to be really worth a dedicated long term campaign effort specifically against you or your company. If you're doing controversial/novel research, are managing millions of dollars, performing a service a state actor would object to, or have high profile clientele then maybe you fall into this category. These are patient campaigns and want to make the domain "feel normal and official". They won't do anything public with the domain such as SEO tweaking or link spam, they'll use these domains only for specific targeted one-off low-noise attacks. They're relying on staff to see that the domain has been connected to your service for years and is likely just a domain someone in marketing purchased and forgot about. This is exceptionally rare.

Wake up. Feed the dog. Make coffee. Take the dog for a walk (~30 minutes), usually put on a podcast.

Check my calendar so I have an idea what my schedule looks like for the day, especially looking for start-of-day meetings. Start "spinning up" in my head. Shower & clean up, dress for the day.

If I have spare time before my workday starts I'll usually spend it catching up on personal email or checking the news. I very specifically avoid "fun traps" in the morning, no personal projects, books I may be hooked on, entertainment media, or engagement based platforms I might loose time to.

At the end of the day I cap it off with a proper workout and another walk with the pip. I find having a semi-enforced schedule to actually leave the house at the beginning and end of the day give me good mental touchstones to know when I should/shouldn't be working. I still find myself occasionally picking at something in the evening, but at least when I do I'm making an explicit choice to _go back to work_.

This is actually the first project I've seen that is actually doing any kind of knowledge graph generation. Most are just precomputing similarity scores as edges between document snippets that act as their nodes. People have basically been calling their vector databases with an index a knowledge graph.

This is actually attempting fact extraction into an ontology so you can reason over this instead of reasoning in the LLM.

Prompts are a great place to look for these, but the part you linked too isn't very important for knowledge graph generation. It is doing an initial semantic breakdown into more manageable chunks. The actual entity and fact extraction that actually turns this into a knowledge graph is this one:

https://github.com/OpenSPG/KAG/blob/master/kag/builder/promp...

GraphRAG and a lot of the semantic indexes are simply vector database with pre-computed similarity edges which does not allow you to perform any reasoning over (the definition and intention of a knowledge graph).

This is probably worth looking at, its the first opensource project I've seen that is actually using LLMs to generate knowledge graphs. This does look pretty primitive for that task but it might be a useful reference for others going down this road.

GraphRAG isn't quite a knowledge graph. It is a graph of document snippets with semantic relations but is not doing fact extraction nor can you do any reasoning over the structure itself.

This is a common issue I've seen from LLM projects that only kind-of understand what is going on here and try and separate their vector database w/ semantic edge information into something that has a formal name.

Don't count it out yet as being problematic for software engineering, bu not in the way you probably intend with your comment.

Where I see software companies using it most is as a replacement for interns and junior devs. That replacement means we're not training up the next generation to be the senior or expert engineers with real world experience. The industry will feel that badly at some point unless it gets turned around.

Combinatorial can be quickly calculated primarily using factorials. 4 possible options, each where you're picking all 4 exactly once is 4!. The reasoning is pretty intuitive, when you start selecting there are four options, when you go to pick the next one there are 3 left in the pool, then 2, and finally 1. This turns into 4 * 3 * 2 * 1 = 24.

This site seems to have a pretty good overview of them if you'd like to become more familiar: https://www.geeksforgeeks.org/mathematics-combinatorics-basi...

That isn't really the case here. It didn't start with a company. The company came along independently, effectively took control of an open source project that was already well established. They found that support alone wasn't enough to be profitable on and have been lashing out at the community that built around the original project, including independent opensource clients.

It's not that the company shouldn't have the code as open source. The company shouldn't exist in the first place and we're all suffering for it.

Don't do this. Don't suggest its safe for others to do. It's VERY not safe to bring anything ferrous or magnetic into the presence of an MRI. You may destroy the device that is going to help save someone's life.

No, but my work is holding my paycheck hostage.

So many unfortunately don't have a choice

Courts are political entities but this is one that Israel chose to accept and recognize the authority of. It has a history of being very transparent in its decisions and is widely recognized as being neutral and fair in their decision making process.

Of course the person charged and found guilty of a crime will argue against the court. Disagreement, even if valid, doesn't change the recognized authority of this court even if the "teeth" are extremely limited.

I disagree with your distinction on the environments but understand your argument. Production for VM to me is "on the road when a customer is using your product as intended". Using the same artifact for those different environments isn't the same as "running that in production".

Not quite. VW got in trouble for running _different_ software in test vs prod. These optimizations are all going to "prod" but are only useful for specific targets (a specific game in this case).

Awww I got really excited but it's limited to Copilot chat.

Anecdotally, I've stopped using OpenAI entirely for coding activities in preference for Claude due to the _vastly_ improved performance on my coding tasks. This was before the recent update to Claude and I haven't gotten a feel for how the new version is doing.

Also entirely anecdotally, the newer multi-modal features added to the OpenAI models have _seemed_ to significantly degrade its other capabilities especially coding in languages other than Python and TypeScript and _seems_ to be more repetitive in its answers (likely to get stuck repeating the same incorrect information even after a correction). This could absolutely be a sampling or task bias so your mileage may vary.

I've still found Github Copilot to be useful for VERY SHORT look-ahead/completion but it has almost always assumed too much in the very wrong direction for more than about a line. I haven't tried the Claude version of Copilot but I'm absolutely switching over to it.

Wanted to call out the specific requirements for what the ACM wanted out of their participation in creating a core body of knowledge (from the linked reasoning):

    * It must reflect actual achievable good practice that ensures quality consistent with the stated interest; it is not that following such practices are guaranteed to produce perfect software systems, but rather that doing so can provide reasonably intuitive expectations of quality.
    * It must delineate roles among the participants in a software project.
    * It must identify the differential expertise of specialties within software engineering.
    * It must command the respect of the community.
    * It must embrace change in each and every dimension of its definition; that is, it must be associated with a robust process for ensuring that it is continually updated to account for the rapid change both in knowledge in software engineering and also in the underlying technologies.

It then details exactly how SWEBOK fails to meet those (which all still seem to be relevant) and comes to the following scathing conclusion:

    Overall, it is clear that the SWEBOK effort is structurally unable to satisfy any substantial set of

the requirements we identified for bodies of knowledge in software engineering, independent of its specific content.

I haven't read the SWEBOK but some spot checking and a review of the ToC seems to indicate they have not meaningfully taken that criticism into an account.

That line is referring to shared libraries linked to a dynamic executable. If a shared library isn't installed or available you will receive an error similar to the following:

    $ ./main
    ./main: error while loading shared libraries: librandom.so: cannot open shared object file: No such file or directory 

Which is indeed a runtime error.

There is also the common use case of hot-reloading compiled code which dynamically loads and unloads shared libraries without restarting the entire application. It needs to be done carefully but you've likely used an application that does this. Failure to load a library, or loading an incompatible one will also create a runtime error.

Looks like there is a lot of bad generalizations in here, but they are technically correct on this one.

This is both uncontroversial and incredibly spicy. I love it.

Just a tactful reference hahah

> the US government isn’t burning taxpayer dollars on a ten figure spaceship just to have us push a Crowdstrike update on it.

Seconding the vxWorks and bare metal. Never seen Windows or Linux on a satellite bus. Haven't really touched payloads but I've seen some wonky things shipped to orbit by universities and not all them have been cubesat student projects.

There is definitely a specific monetary threshold where the financial justification flips and its probably around about $1 million/year of cloud costs if you need two region equivalent. This isn't "build a data center" kind of money, this is "rent some controlled racks and hire staff". You'll have to eat a good chunk of that operational savings over the time window with a big upfront capital hardware purchase.

That being said, this reeks of making a anti-monopoly argument and seems wildly overstated.

This is the full or classic version of clean room reverse engineering. Using unencumbered public documentation is relatively new, that kind of detailed documentation wasn't widely available. Car manufacturers still protect their service manuals with an agreement that basically says they can't be used for this but I think a lot of service centers stopped making people sign them.

The classic tech story that used this technique is the IBM BIOS and the resulting spread of "IBM PC-Compatible" machines. There is a little bit about it on the wikipedia page (https://en.wikipedia.org/wiki/IBM_PC%E2%80%93compatible). Random factoid, the Netflix Original "Halt and Catch Fire" has a depiction of doing this IBM clone reverse engineering and did a pretty good job at it.