> Latency adds up. Each delegation is a network round-trip
not a big expert on agentic coding, but why on earth the delegation needs to be to a remote something? can't this be a subagent/MCP server/tool running on the dev machine?
Wouldn't moving the code elsewhere create a double working copy, possibly different for the current edits?
The GPUs and the people (I imagine you mean the extremely well paid researchers) are there because of the funding.
The funding is in the US now, but as TFA explicitly says, the law would reduce spending on AI training. All the money will still be available, which means the funding will either a) move to a different sector or (more likely) b) move to a different country. Taking the people and the GPUs with it. So yes, most training will move offshore.
Secondarily, OSS business models are based on the fact that the developer is the most qualified in offering support.
With models, anyone with the weights can serve the model with the same performance (meaning level of "intelligence") as the original developer. Any business model depends entirely on spending huge amounts of money in training to then hopefully gain from inference. If anyone can compete on inference the same day you deploy the model, it's unlikely you can break even while your competitors that saved on training get rich on inference. i.e. it's more likely to entirely stop training in the US rather than slowing it down...
According to wikipedia, EU has a 23 trillion GDP (30 trillion PPP) and 451 million people, making ~51k GDP per person (67k PPP).
US is at 32 trillion GDP and same PPP for 342 million people - 94k per person
code with compilation instructions can be understood and used by much more people than a long conversation between a human and 2 top tier LLM only accessible through costly agreements and providing nondeterministic output. Going from a prompt to working code you have to traverse a pretty steep paywall...(yes 20 to 200 dollars per month is steep if you compare it with the rest of the software stack like compilers/interpreters and text editors, which is usually free)
I think the fact they are known to have changed the content of archived pages is way worse than the DDoS incident, since it makes them untrustworthy even as a source of information. Wikipedia decided to ban all archive.today (and aliases) links because of that
What I fear is Trump didn't break any promise done to their electors by starting this trade wars (he did when he failed to stop actual military wars and started more instead, but that's a different point). He was elected by stating that America was the greatest country in the world and that the others were freeriding on US wealth. I'm not sure his successor can entirely overturn this policy, if a majority of the US voters believe this - also because undoing the damage would require brokering deals less favourable than before to show good faith.
I live outside of the US so I might be misintepreting the situation, but I'm unconviced that a full u-turn will happen once he's out of the office.
pivoting is useful if you are doing several things, one well and the others badly. You stop doing whatever it is you are bad at and focus on what you are good.
If your company is going strongly overall, you can just continue doing whatever you are doing (adapting to the market is still necessary, but not with huge changes)
TFA surely refers to management consultants, your parent looks like it as well (when they talk about management obsession with consultants).
Technical consultants are mostly OK if their skill is not your main line of business, you just need a few of them or you need them temporarily. Companies employing thousands of software engineers or other professionals from consultancy firms only demonstrate their inability to plan, hire and organize people. It usually correlates with companies employing swats of management consultants.
if you take a model that requires 200GB of VRAM and you run it on the CPU, it requires 200GB of RAM instead. Still unfeasible on consumer hardware. With this approach you can easily do it on 12GB or less of either RAM of VRAM, at several seconds per token instead of tokens per seconds. Very unusable, but certainly interesting!
I think that's not capitalism, that's human nature. On every possible power structure you can think of, there will be people available to compromise on their moral principles if given the right incentives to do so. Under capitalism that's easily money, under other systems it might be something else. You can work to mitigate or detect and deter it (the amount of effort put on that is what makes the difference between limited and widespread corruption). You cannot hope to eliminate it entirely.
do consider that in Italy most credit cards cost around ~30€ per year + a 2€ tax for all months you spend more than ~70€ and most offer no benefits. Debit cards are offered for free by all banks. So credit is only used to rent cars (not really mandatory anymore) and if you really need the credit (but other ways of getting short term credit exist now).
I think one of the points of TFA was that other AI tools found many vulnerabilities; after having fixed those, mythos did find another vulnerability the others missed, but that seems to imply this model is only marginally better than the competition instead of being on a different league altogether like it's marketed.
Paraphrasing the author: sure mythos will find lots of security issues in gnutls, but so will gpt or opus (they acknowledge explicitly that all those tools are getting very good).
> Yesterday, I read a Wikipedia page for a book I’m about to review. I am still unsettled.
> The page was stripped of reality, and in its place was a sanitized fairytale where Putin is good and the book — a brutal and damning historic account of Soviet abuses — is subtly and not so subtly undermined from every direction.
We have no idea which book that was and what was so unsettling about that entry
I did not say they did not read the title of the article. They clearly did. It's the rest of the content that was lost, such as the long focus on the sloppiness of the API provider
The article author did not even bother to read the article they were basically replying to. Otherwise he would have noticed that the main points the OP was complaining about were not about the agent, but the hosting provider providing an API allowing destructive operations easily, using tokens with no scopes, with backups stored in the same volume as main data, etc. So this article is actually agreeing with the complaints of the original article, just more generically and without spending an effort on it, doing that with a tone that implies the original article writer is an idiot.
In general, with per minute rate limiting you limit load spikes, and load spikes are what you pay for: they force you to ramp up your capacity, and usually you are then slow to ramp down to avoid paying the ramp up cost too many times. A VM might boot relatively fast, but loading a large model into GPU memory takes time.
IANAL, but I think open source started with software since software has source and binary form. Now with compression and other shenanigans, probably even videos or images could be argued to have a source and binary form. I don't know a thing about multimedia, but people here saying this "open source" release is a good thing mention specifically the fact that it's the uncompressed version, or as the FSF would call it, "the preferred form of the work for making modifications to it".
The guy is actually the maintainer of those packages. So whoever got his credentials became able to perform releases on those packages. NPM itself does not build any package, it's just a place where people can publish stuff
Yes, you can have Android apps showing up in Gnome (e.g., likely all DEs) menu and open them directly in their own window without seeing the waydroid launcher.
I think it's a bit more subtle than that. The code of this tool runs in your browser and makes it download the model from huggingface. So it does not host the model or provide it to you, it just does the download on your behalf directly from where the owner of the model put it. The author of this tool is not providing the model to you, just automating the download for you.
Not saying it's not a copyright violation, and IANAL, but it's not a obvious one.
> Besides, can we even attract experienced developers to a non-glamorous industry like logistics?
I wouldn't worry about this. For some devs the industry might matter, but for many code is code, independently if it's for finance, healthcare or logistics. Just make sure salaries are in line with the market.
Internalizing devs will likely produce cost savings long term unless off the shelf tools are good enough for you, but it's an investment that needs long term commitment. Initially you'll still have to pay for your existing product plus the devs, so it'll feel like you just added a cost. Make sure you are committed enough to pass that phase. Your first hires should be able to work independently, talking directly with employees about requirements, without requiring extensive documentation or dedicated PMs before starting. Too much structure initially would be detrimental, you'll probably want to keep the team small anyway.
They are quite well differentiating generative AI models, that are not trained on customer data because they could leak customer data, and other types of AI models (e.g. recommendation systems) that do not work by reproducing content.
The examples in the linked page are quite informative of the use cases that do use customer data.
You and many others here are conflating AI and LLMs. They are not the same thing. LLMs are a type of AI model that produce content, and if trained on customer data would gladly replicate them. However, the TOS explictly say that Generative AI models (=LLMs) are taken off the shelf and not retrained on customer data.
Before LLMs exploded a year and a half ago lots of other AI models had been in place for several years in a lot of systems, handling categorization, search results ranking, etc. As they do not generate text or other content, they cannot leak data. The linked FAQ provides several examples of features based on such models, which are not LLMs: for example, they use customer data to determine a good emoji to suggest for reacting to a message based on the sentiment of the message. An emoji suggestion clearly has no potential of leaking customer data.
well, you need to have something to run that business logic (which includes phoning home to the manufacturer), don't you? Java is as good as any other runtime.
> Periodical reminder that copyright and GPL do not talk about linking.
> At all.
GPLv2 does not, but both LGPL and GPLv3 explicitly mention linking in the license text, so this part of your comment is factually false.
However, it's mostly used as an example; I agree any work with strict dependencies to a GPLed work might be considered a derivative even if the dependency is not expressed by linking.
> We ignore their "we noticed you may benefit from Enterprise, call us today!" spam
Of course if you ignore their offers to talk with the sales team, you won't talk to them. Bigger companies than yours will be interested in their enterprise package and/or will want to negotiate volume discounts.