It's all software in the end. If you plan ahead, it is hard to be completely locked in. I worked for a company, where we had build a data transformation pipeline, and we had GCP and AWS credits, and wanted to use both of them. We built a minimal, high-level library that could be mapped onto both PubSub and SQS (we were transforming the WARC files from the internet archive, in preparation for data analysis). Once we built this, and worked around the near-criminally incompetent design of SQS (e.g. PubSub does compression, batching, and does not send duplicate messages, while you need to all that yourself for SQS), we were able to use both clouds (for free) at the same time, and we were not locked in at all.
In fact, we had a call with some people who worked at GCP, and I (naively, and excitedly) mentioned that our design is simple/minimal and cloud-portable. A miasma of awkwardness fell upon the conversation, and I realized that the reason that the GCP people were so keen to speak to us, was that they were anticipating, that once we used up the credits, that they would then get regular recurring revenue from the company.
Most cloud companies are in the business of locking you in via very convenient APIs, that would be costly to migrate off of, if you did not abstract them away to begin with (naturally, the egress costs cannot be abstracted away, but those were not much of a problem in our case, because our output-data was very small compared to the input-data).
There is an expression in my native language: he who lacks virtue in the head, has virtue in the legs.
If you scroll down, most of the charts on that page are sorted s.t. Mistral's bar is right next to the worst competitor model, while the best competitor model's bar is positioned on the opposite side.
If one were to be cynical one could say that it's intentionally making Mistral's result look better than it actually is by making it harder to compare the bar heights.
Bar charts should start at zero. If they don't start at zero, there should be a clear visual indicator that the chart has been trimmed without having to read the axis labels. I hate that this has to be repeated so often that it has become a cliché.
That very blog post describes where they got their materials from, most of it comes from Retromags and out-of-print-archive. You can find downloads over at their sites, and those sites usually also upload to IA.
I don't think you can download directly from VGHF as that is not what it's intended for (it's a research tool)
no idea who Frank is (neither do I care) but these files have been readily available for years (and still are), all they're doing is collecting them at a central place and make them searchable (and vievable at a decent speed, compared to IA).
Preservation through proliferation. Should the archive be forced offline, for whatever reason, it will still persist elsewhere, even if its just rotting on somebody's external hard drive, waiting to be found.
Can't speak for the above, but for me it's a hedge against enshittification and sudden shutdowns. If I had a dollar for every community archive that suddenly went offline, I'd have several dollars.
This is the archive of the Video Game History Foundation. They’re properly setup legally, have overwhelming backing from the industry, and a pristine reputation. They’re basically the Smithsonian, or more accurately the Isabella Stewart Gardner of the video game world.
No. This is by design. I do not like it. It's a tactic often deployed by libraries that want to retain some control over scanned assets. To be fair, normally they are funded by tax payers.
They usually roll things out in the EU after a few days/weeks. It's too juicy of a market to ignore for them. It's an easy way to phase in a new product without having to commit to world wide coverage on day 1.
Plenty folks in Europe work for American businesses, including tech, it’s just super duper hard to scale a business as quick as it is in the US for a variety of reasons - regulation, funding, labor laws, euro idea of work life balance, …
Fragmented EU domestic market (27 very different countries with no shared language) vs. unified US domestic market (50 pretty similar states with shared language) is the most important factor when it comes to scaling the typical services with marginal to nonexistent unit costs that are common in the tech world, especially consumer tech.
Yeah but also the consumer (if you’re selling to consumers) is very different in the US va Europe: on average, the US consumer wants to be sold stuff to. Not as much in EU.
Thank you. People love to blame regulation but no amount of regulatory simplification is going to make a Lithuanian speak Italian.
Combine it with the fact that in the media of every country the US comes far ahead of any other UE countries in terms of coverage (on any topic), you have a recipe that no amount of deregulation can ever fix.
Whatever LLMs are capable to with regard to translations is not relevant for the 30-year history of primarily Silicon-Valley-originated consumer tech service expansion: todays' LLMs didn't exist pre-2022, but the gigantic moat of US tech giants was developed pre-2022. Today, they mostly profit from that flywheel having been sped up already in the past.
Why should they join in on a shoveling cash in to the fire competition? The winning move here is to just sit and watch the others waste all of their resource and to just take the free open source model that comes out at the end.
Again; they don't need to. Chinese open-weight LLMs are free and better than US models for security-focused work.
As an American, I wish our country had fewer "frontier" labs. They are not profitable and will destroy the economy when (not if) they fail to achieve gross ROI.
I kid you not, the research cloud for Danish universities is deploying a "sovereign AI". Which means running GLM 5.3 on their Nvidia GPUs. All that while EU designates China a strategic threat.
That is unironically more sovereign than using any US-developed AI. GLM 5.3 is cutting-edge for cybersecurity tasks, while the US frontier forces KYC/background checks for cyber capabilities.
Well in the same way as all the solar panels come from China, as well as all batteries.
EU is an open market - it exports a lot of products, but they are of high value. The AI spending is simply not high enough value so that it is picked up.
Well it's too late now. Europe should have started in 2023 when there were only a couple of models. Instead, we got Mistral, which is just useful for OCR.
A technology whose main selling point is that it will leave you without a job? What's wrong with Europeans, why don't they love this? Maybe because we don't have enough billionaires to convince us.
reply