estebarb

It feels like enforcing that everything must be written in Ada. And then relaxing the requirement... history always repeats itself, it seems.

Thousands of years ago there was a breakthrough discovery that shaped humanity forever: living in society.

Do we need to explain that one of the perks of society is pushing others forward with a tacit expectation that it will come back for everyone eventually?

In the post-truth era, with fascism gaining adepts all across the world... who would want to give a government editorial powers on generative AI?

I'm deeply pessimistic.

You know that in most countries the taxes must be shown explicitly and be added in the final price? USA is the only exception.

Without fair trial there is no proof they are associated with gangs and violence. Thousands of innocents are jailed in there.

https://www.rfi.fr/es/programas/noticias-de-am%C3%A9rica/202...

There are already words for that: banished, disappeared, forced exiled, concentration camp victim... just reuse terms already used to describe crimes done by nazis and other fascist goverments.

The default where? In Windows the default is Edge, in Mac is Safari and in Linux distros usually is Firefox. And somehow people prefer to download Chrome.

By hardwiring the allocator you may end up with binaries that load two different allocators. It is too fun to debug a program that is using jemalloc free to release memory allocated by glibc. Unless you know what you are doing, it is better to leave it as is.

I think we would have preferred continuing with .NET, as no one is Rust expert on the team. But binary size and lack of some SIMD instructions moved the balance to Rust. And then, the PoC had big memory usage improvements, so...

Sadly yes, we have data. We are migrating our C# SDK to Rust in part because customers want a much smaller dependency. And the AoT compiler didn't trimmed as much as we wanted.

I have no idea about EBM, but I have researched a bit on the language modelling side. And let's be honest, GPT is not the best learner we can create right now (ourselves). GPT needs far more data and energy than a human, so clearly there is a better architecture somewhere waiting to be discovered.

Attention works, yes. But it is not naturally plausible at all. We don't do quadratic comparisons across a whole book or need to see thousands of samples to understand.

Personally I think that in the future recursive architectures and test time training will have a better chance long term than current full attention.

Also, I think that OpenAI biggest contribution is demostrating that reasoning like behaviors can emerge from really good language modelling.

The problem is exactly that: the probability distribution. The network has no way to say: 0% everyone, this is non sense, backtrack everything.

Other architectures, like energy based models or bayesian ones can assess uncertainty. Transformers simply cannot do it (yet). Yes, there are ways to do it, but we are already spending millions to get coherent phrases, few ones will burn billions to train a model that can do that kind of assessments.

C# AOT filesizes are huge compared to Go.

I believe that it is related with important conferences opening papers reception soon. Some disallow publishing in preprint for some weeks before the paper reception, so people may have been rushing uploading stuff.

I'm not sure about it makes sense to apply Gödel's theorem to AI. Personally, I prefer to think about it in terms of basic computability theory:

We think, that is a fact.

Therefore, there is a function capable of transforming information into "thinked information", or what we usually call reasoning. We know that function exists, because we ourselves are an example of such function.

Now, the question is: can we create a smaller function capable of performing the same feat?

If we assume that that function is computable in the Turing sense then, kinda yes, there are an infinite number of turing machines that given enough time will be able to produce the expected results. Basically we need to find something between our own brain and the Kolmogorov complexity limit. That lower bound is not computable, but given that my cats understands when we are discussing to take them to the vet then... maybe we don't really need a full sized human brain for language understanding.

We can run Turing machines ourselves, so we are at least Turing equivalent machines.

Now, the question is: are we at most just Turing machines or something else? If we are something else, then our own CoT won't be computable, no matter how much scale we throw at it. But if we are then it is just matter of time until we can replicate ourselves.

Actually, "hiper-something" trend started with Hipermas, created by a bigger local supermarket chain (Corporación Supermercados Unidos, which owned others like: Más x Menos [More for less] and Pali). They were bought by and renamed to Walmart. Sharing the "hiper" prefix I only know HiperDiego, which like a Walmart big.

Here most are "super-" if they are big enough. if they are smaller "mini super-". All the categories before are self service.

If they are really tiny (a window) and not self service then they are called "pulpería".

So, we will see "America Mexicana" in Google Maps?

There are tricks: if you need to find uniformly distributed data it is unbeatable: guaranteed to work in 5 hops or less. After 5 hops, is better to use binary search. If data is not uniformly distributed one trick is to sort by hash of the element (store the hash with the element or calculate on the fly). But yeah, without those adjustments it has a worst case of O(n).

Sure, but those developers will create functionality that will require advanced GPUs and people will want that functionality. Eventually OS will expect it and it will became default everywhere. So, it is an important step that will push nvidia growing in the following years.

Recently I tried to optimize a set implementation and found that interpolation search works quite well, being competitive with HashSet in C# (both single digit ns for my use case), with zero memory overhead. Unfortunately, it only works well with uniform data, so I guess I'll give static search trees a try. This explanation was clear and extremelly well detailed.

Clearly it is Big Compiler propaganda! </bazinga>

The problem is private insurance companies + private hospitals. They shouldn't exist.

Healthcare is an inelastic service: people will pay everything to get it. So private hospitals in a free market are pushed to bill their services as higher as possible. Actually, their prices would go to infinity, if not limited by people total savings and earnings.

So, here the insurance companies make the problem worse: they give people access to "infinite" credit. So you can pay for those exorbitant prices. But with insurance you just increased the pool of money people can use for health, so private hospitals can and will increase their pricing. Add more free market competition and you get the disaster of the USA healthcare system.

Because of that reason, basic economics, is that for profit Healthcare cannot work long term. The only way to make it work is making it a public service. But in USA politicians will cry that is comunism, so they won't do it.

And the USA Healthcare problem doesn't limit itself to USA. As americans are unable to pay for healthcare there, they started doing medical tourism, which is making health prices in other countries more expensive too.

At the university squirrels regularly stole our pizzas ( see https://imgur.com/gallery/HBLFoNA ). At my house, they also took avocados.

Yes, exactly. So, this is a huge security gap.

As an attacker, instead of DDoSing a service we could just upload a bunch of NSFW text so Google kills their infra for us.

Other providers, like OpenAI, at least provide a free moderation API. Google has a moderation API, that after the free 50k requests it is more expensive than Gemini 1.5 flash (Moderation API costs $0.0005/100 characters vs Gemini 1.5 flash $0.000001875/100 characters).

I asked Gemini about banning risks, and it answered:

Gemini: Yes, there is a potential risk of your Google account being suspended if your SaaS is used to process inappropriate content, even if you use Gemini to reject the request. While Gemini can help you filter and identify harmful content, it's not a foolproof solution.

Here are some additional measures you can take to protect your account:

* Content moderation: Implement a robust content moderation system to filter out inappropriate content before it reaches Gemini. This can include keyword-based filtering, machine learning models, and human review.

...

* Regularly review usage: Monitor your usage of Gemini to identify any suspicious activity.

* Follow Google's terms of service: Make sure that your use of Gemini complies with Google's terms of service.

By taking these steps, you can minimize the risk of your account being suspended and ensure that your SaaS is used responsibly.

---

In a follow up question I asked about how to implement robust content moderation and it suggested humans reviewing each message...

For me that is a reason for not touching anything from Google for building stuff. I can afford lossing my Amazon account, but Google's one would be too much. At least they should be clear in their terms that getting banned at cloud doesn't mean getting banned from Gmail/Docs/Photos...

Kids born in "native english" change the language too. For example, treat irregular verbs like regular ones. Languages evolve with time, regardless of non-native speakers using it.

For those scenarios would be helpful a draft generation mode: 16 colors, 320x200...

Except that it is not. If you want to use this for tasks like navigation then it is useless. Practically, it is the same as using chatgpt to generate a realistic scenery based on a text description.

Something that concerns me the most is how are we going to train new generations. I teached a course at the university and many students just chatgpted everything, without any critical thinking.

It doesn't matter how many times you showed that it invented assembly instructions or wxwidgets functions, they insist on cheating. I even told them the analogy of going to the gym: you lift with your own strength, you don't use a crane.

And of course, it is evident when you receive students that don't know what is a function or that cannot complete simple exercises during a written test.

We learned by reading, lots of trial and failing, curiosity, asking around, building a minimal reproducible bug for stackoverflow... they (the ones that rely only on chatgpt and not their brain) cannot even formulate a question by themselves.