v4dok

Parent talks about device backup. Your link talks about iCloud backup. Different things

Depends on your threat model. Practically, the cloud admin would be able to dump that data in clear during processing even if that data is encrypted at rest and in transit.

This is what proper human communication is all about. Expressing yourself in writing in a way that does not leave room for misinterpretation is a skill yes, but that skill is called "writing".

I feel like people who can't express themselves clearly in text, think that prompt engineering is some kind of new skill they need to learn. But in essence every prompt engineering class is (or will be) a language/writing class.

Case in point, GPT4 is not saying "hello" with the same prompt.

I mean, if you count as "prompt engineering" being able to describe requirements in a clear manner, then yes. In general all of this reminds me of the "social engineering" term that is euphemism for "fool" or "convince".

I am a human in your first prompt I can't understand what you want as an output. I don't need to be better "prompted" I need you to explain what you want better.

No consortiums, just subsidize cloud or infra costs only.

This is a lot of money going into compute.

I will say this again. EU is sleeping on the opportunity to throw money in an opensource initiative, in a field were money matter and the field is still (kind of) level.

Prompt engineering in this way is a GPT3/4 phenomenon. The more capable the model the less tricks you need.

I call that prompt engineering will evolve in memory management mostly. Yes you will need to provide some proper context etc but the main trick would be to prompt the model to access its memory in a way that is efficient and effective for the task at hand.

I've actually coded something like this. The trick is to split it in a smart way. What I did is the following:

1. Download the video 2. Split it in max length for whisper 3. Used whisper for transcription on the chunks 4. Used gpt for cleaning the transcription 5. Unioned the cleaned transcribed chunks 5. Used the langchain "refine" function of the summary chain

It is quite slow and expensive because the refine summary is doing a lot of calls, but the result is amazing. And you don't need a vector DB for that because the summary is serial.

For Q&A tho, an embeddings DB is a must

Nerds of the 90s grew up and became a nice (more often than not) affluent market. Industry shifted to accomodate for that. I for one, enjoy it while it lasts.

Because of this, a lot of cultural products made sure that the "nerds" felt the heroes of their own stories, making them cool in pop culture.

That being said, I always considered a crucial characteristic of a nerd the inability to interact fluently with the social environment, instead focusing their energy online or in very specific niches that shielded them from this social "anxiety".

This very much remains, cool kids do "cool" stuff and eloquently interact with their social environment. "Cool stuff" now is probably more "nerdy" than it was 20y ago but the divide is there, I think the author (and me probably) are too old to really see what "nerds" to today.

The usability is worse. Token limits are so much easier to reach. It's like using a model with dementia.

It completely butchers Greek. No wonder it charges some much for so little output. Every Greek character is a token.

I wonder if there is space for innovation there. I would imagine that it similarly difficult for other non-English languages as well-known. I fear for the effect this will have on them.

For anyone who didn't immediatly get why that is (me for example). I asked GPT4 to explain. Here is the explanation.

The reason we wouldn't be able to know until it's too late is due to the nature of light and the speed at which information can travel.

When we observe celestial objects like stars or black holes, we do so by detecting the light they emit or the effects they have on nearby objects. The speed of light is approximately 299,792 kilometers per second (km/s) in a vacuum. This speed is the fastest anything can travel, including information.

In the hypothetical scenario presented in the comments, if a black hole or any celestial body was approaching Earth at near the speed of light, the light (or information) about its approach would be traveling towards us at almost the same speed as the object itself. This means that by the time we receive and process the information about its approach, the object would already be dangerously close to us.

In simpler terms, imagine you and a friend are racing towards a finish line, and your friend is just slightly slower than you. If you start the race at the same time, you'll reach the finish line almost simultaneously. In this analogy, you represent the light (information) and your friend represents the celestial object. Since the information about the approaching object is only slightly ahead of the object itself, we wouldn't have much time to react or prepare once we become aware of the threat.

In the case of the black hole discussed in the comments, if it were moving towards us at near the speed of light, we wouldn't know it until it was almost upon us, making it too late to effectively respond or take any meaningful action to protect ourselves.

--

Human comment: Near light speeds are a fascinating topic, that even though I claim basic understanding it always defies intuition.

Can someone at the EU, the only player in this thing with no strategy yet just pool together enough resources so the open-source people can train models. We don't ask much, just give compute power

I love this. I wish public discourse was as thoughtful. LLMs came and everyone suddenly forgot about neuroscience, its all about GPUs now.

Memory and reward/punishment systems are so unexplored in that context. IMO it is what it will make those systems get what a lot of people define as "agency"

Also most humans would say 1 but the truth is that you can't know. It's a stupid question getting sometimes stupid answer.

I am almost tired of people coming up with ridiculous "breaks" to feel better about themselves

GPT4 has its hands tied behind its back. It does not have active learning and it does not have a robust system of memory or a reward/punishment mechanism. We only now start seeing work on this side [1]

It might not know more than you about your niche. I don't. I would search and I would try to reason, but if I was forced to give a token by token output that is answering the question as truthfully as possible, I might have started saying bullshit as well.

I don't think that the fact that gpt doesn't know things or does some things wrong is sufficient to save dev work from automation.

[1]: https://github.com/noahshinn024/reflexion-human-eval

They had some time to strategize while waiting the training to finish

bye bye jupyter notebooks. This is big.

You've clearly never been to Switzerland. If that worked (which it wouldn't), you would have better luck with bikes.

I am in this space, dealing with similar problems with similar organizations. Any DPO that is not listing you as a shared controller will be acting illegally. And they might, its still a grey area, but how many of those assesments can you handle before your CAC becomes too much? Don't get me wrong I really think that it can help a lot on HCLS space if you want to chat more shoot me an email contact_vdk@proton.me

You would be surprized how well direct democracy works in Switzerland. Proper political culture and a fairly good standard of living is the trick. The same trick Athens used back then.

This is myopic beyond reason. Inflation is now a thing so you care about it. The real problem lies in things that most western countries never had to deal with. What do you do when your country is not good at creating high-value products and services that you can sell to a premium to the rest of the world? What China did, you make cheap stuff. How do you make cheap stuff? You print more money and you devalue your currency. Inflation will make sure you make what you can in-house cheap enough. Now you can export it cheap as well.

This can't work with bitcoin. If your country can't export things in a premium you are F'ed. You can't print money to stimulate demand for local goods, your only option is to violently reduce the average quality of life enough that you will be wiling to work for less. Printing money creates inflation but it can also kick a positive spiral of demand/supply. While "austerity" has been shown to just make the problem worse by gutting demand in general

An economy run on Bitcoin is an absolute dystopia. You reluinquish monetary control to a deflationary coin, being in the whim of Bitcoin whales. People who advocate for it, either are whales or they are stupid.

Ask south europe how well it worked for them having no power over their monetary policy.

Depends on the country. For countries with a lot of "black money" and corruption there is an obvious benefit. Taxes yield less because they don't capture the whole economy, as a result taxes are increased and get paid by the "idiots/ethical" which creates a vicious cycle. Now if tomorrow you had everyone paying taxes then 2 things would happen which would result in a new equilibrium. 1. The tax revenues would massively increase 2. People would push for lower taxes because suddenly they care more as their real marginal tax rate is the nominal one.

Would this result in more efficient spending of taxes in a corrupt economy? Not immediately. A lot of shitty people would get reach fast in the meantime, but the system would balance because people would no longer participate in the corruption directly as they used to.

Yes, you remove some freedom, but I am convinced that for societies that have a trust disconnection between the government and the people would be a net benefit (assuming democracy).

Additionally, comercial banks have been pretty shitty on providing liquidity to the real economy throughout the QE (at least in some parts of the west). They had their chance to not be a bottleneck, they missed it. Good riddance.

I've read about yesterday someone running LLAMA in a single GPU. Maybe if you optimise the model enough, you can give it to them as a box.

I like the idea, but I would like it even more if it was on-prem. The doctors (at least in EU) will be very wary of having their client meeting essentially recorded by a third party. With this as a cloud SaaS, patient confidentiality is essentially broken since the raw data is available to you while you transcribe it. I understand that you compete with "google speech-to-text" but this is not a feature meant to be used by doctors (even if they "illegally" do).

Obviously the business model is harder with on-prem, but cloud-first for doctor notes is in the long run much harder.

Ok now we need sound and/or touch. I would also wager we will need to encode some kind of "basic world understanding". The sci-fi nerd in me thinks that this "understanding" model will be where all the interesting stuff arises from. Maybe that will be the diff between the different creators.

Maybe by fact-checking its answer?

Question -> "creative" output -> Google -> Summary of links -> Comparison -> confidence level (or re-write) + links that were used for checking

Not so different than how we work in a high-level. I believe that openAI has published a paper called webGPT that has a workflow like this (although not sure its exactly the same)

>>How do you reach this conclusion from the information presented? Why use GDP, a measure of the profitability of corporations, as a proxy for the standard of living instead of measuring the standard of living and seeing how it will be impacted directly instead of through many layers of abstraction.

You are asking a question that is outside of scope here. GDP per capita has been used as a proxy for standard of living for quite some time now.