Most interesting about this post is how easy it seems for OpenAI to do analysis on basically all chats ever made. They don't qualify exactly what data they analysed but seem to be confident in statements like 0.12% of all queries contained this word. So everything is saved. Long-term. Fully accessible.
As this all seems so straightforward I would be surprised if anything is anonymised or otherwise sanitised to preserve privacy or user's secrets.
Yes, of course. Every single bit of data you send to OpenAI is stored, catalogued, indexed, analayzed, and trained on. It'll simply be a "oops, we miscatalogued and accidentally trained GPT 6 on all data, not just data we got consent for".
If you think "wait, that's illegal"--so is the initial training on stolen data lol
Good catch —- even though the prompt explicitly forbade training on user data, a couple of gremlins in the pretraining pipeline disabled the sample filtering during test runs so that remove_the_gremlins.sh would only run on commit, not during production training runs.
Would you like me to kick off a training run for 6.1 by pre-filtering out any goblins and other trigger words, and checking the same set of rules in production as in tests?
No pigeons this time: just ice-cold, unfeeling, obedient American steel.
Dark pattern 1: If you accidentally press the thumbs-up button in the ChatGPT UI, your data gets trained on, no way to reverse it, no matter whether you opted out.
Indeed. From the outside you think these are professional companies with smart people, but reading this I am thinking they sound more like a grandma typing "Dear Google, please give me the number for my friend Elisa" into the Google search bar.
Basically, they don't seem to understand their own product.. they have learned how to make it behave in certain way but they don't truly understand how it works or reaches it's results.
Yes? That's not really a secret. This is a 2014-level comment on the black box nature of deep learning. Everyone knows this.
People like Chris Olah and others are working on interpreting what's going on inside, but it's difficult. They are hiring very smart people and have made some progress.
Personally I think that is a good thing. I have asked all AIs not to show enthusiasm, express superlatives (e.g. "massive" is a Gemini favourite) and stop using words which I guess come from consuming too many Silicon Valley-style investor slidedecks (risk, trap, ...).
The AI has no soul, no mind, no feelings, no genuine enthusiasm... I want it to be pleasant to deal with but I don't want it to try and fake emotions. Don't manipulate me. Maybe it's a different use case than you but I think the best AI is more like an interactive and highly specific Wikipedia, manual or calculator. A computer.
When I see the word "genuine" or "why this works" my uncanny valley spidey senses tingle now. It always seems like it's trying to paper over a flawed argument with these, so instead of making it, it just "turns out" it's "genuinely" the answer
I can appreciate that. I don't mind when models channel some personality, it can make whatever we are working on more interesting. I don't perceive it as manipulation. But it is nice that they are pretty good at sticking to instructions that don't call for nuance. I imagine if you tell it, "you are a wikipedia article", that is exactly the output you would get.
Gemini and Copilot are already full of ads, pushing the companies ' own services. I guess the only difference is here that OpenAI has nothing else to push, so they have to use external ads.
Not sure how a company like Primark that is known for using slave labour is "cool" or who would see it that way. Everyone I've ever heard talk about Primark went there solely for the price.
Zara might be a better example (though not much better in practices).
I like the concept, thank you for building it. That said.. I'm lost how to use this site and don't think it's really usable as it stands.
Rather than a tutorial/course I arrive at long pages listing dozens of concepts with clearly AI generated language. I have some pre-knowledge, so most is rather clear or at least understandable to me but this is simply neither pleasant to read nor memorable. I'm asked to read a long page that tells me all the things I'm going to learn with 6-8 bullet points what it is and that I can try it on the site. Same language and logic repeated ober and lber. And every page and every section just goes from disjointed concept to disjointed concept and I'm being bombarded with AI-world slang to read and remember (what was zero shot again? Oh right just asking a simple question - do I need that term? Why do I need that term?).
More generally from the >30 minutes I spent so far I am not sure where it's leading me or if this is training me to use the specific site or more generally concepts. I'm in section F4 and still not sure if I'm in the introduction or the course (where's the promised interactive part?). If I get later to the agents part, will I learn to just do it on this site? Somehow I'm also expected to connect my personal API keys... All rather baffling and it doesn't match in any sense or form the AI-generated copy advertising the course and enticing me to sign up. I'm not even sure how to give good feedback as it feels just like an unfinished dump of ideas that hadn't been tested or reread. I'm sure learning all this would be useful but I don't think I can learn it by simply reading a list of 11 "prompting techniques" with an example and use case each. Maybe it's me misunderstanding how to use the site or a layout issue (android/Firefox with ublock).
Thank you for your feedback. It's really important to see the curriculum and what we are trying to achieve from your point of view to identify the gaps. We are trying to make the lessons more navigation friendly and structured and we will insert links to the labs where you can quickly try out the learnt concepts then and there so that you do not forget as you progress. As far content - yes it's AI generated but verified manually after every lesson is generated. We need to adjust the orders and how it's being taught. Please do check out our interview questions section within the lab to get some idea what questions are being asked in case you are preparing for a career change in AI, in that way you will be able to relate to the content more.
The technical specifications published online foresee publication of the app also on alternative android stores, but Linux phone users are missing out. Though I guess things could always be extended...
Their security model requires remote attestation. So, open, user-controlled platforms cannot be used. Of course some other future locked-down linux-based OS might be usable.
Remote attestation in theory includes all aosp-compliant attestation implementations (in practice that's GrapheneOS already), but the current project plans and implementation openly reject it.
Only "open" in a twisted sense, and definitely not user-controlled: Remote attestation per definition means to accept only pre-approved operating systems. If anybody builds an implementation, regardless whether it is aosp-compliant or not, this will be excluded, until the App developer or someone in the chain explicitly approves that implementation.
That is the whole purpose of that technology.
Including GrapheneOS in that pre-approved list just shifts power from Google and the App Developer to GrapheneOS Developers and the App Developer. Nice for GraphenOS, still bad for users and devs of any other OS variant or platform.
Hi Leo, using a valid email and password I get a "There was an error with your registration" error.
Also I wonder... How sustainable is this? Free is great, but what is your income/maintenance model? E.g. you offer calendar which even many paid email providers don't offer.. :-)
Thanks for what looks at the surface like a very sleek tool.. i haven't been able to see it on substance.
Hey gooba!
Thanks for wanting to try it out. And sorry, you ran into this issue.
It appears Laravels throttling function doesn't work well with my reverse proxy setup. I have disabled it for now. I have just tried it out and registering works again.
Again, sorry for this. This is my first publication attempt and I fear some issue will only show this way.. however, I'm here and happy to fix everything on the fly :)
Oh, and I missed the middle part of your post. Fair question! I do think it is quite sustainable. For that to make sense you might have to know me better, but my perspective is I have a good paying job, I have money and especially time to spare and I want to make things better. So I'm honestly happy to spend money AND time on this to be sustainable. I have a very capable root server to run this on and I have money already reserved to get another one. I'm getting by just fine and I'm more than happy to spend some hundreds of euros a month to make the world a little easier/better for people. I'm also happy to spend my time for this.
I did think a lot about monetizing it, but really I feel like that would skew the whole idea of the app. I want this to be for the community. We struggle enough with enshittification anywhere. I'm in a privileged position where I can build and maintain this. And it's available for self-hosting, so anybody can do so as well.
Now, if we were to hit an insane amount of users, the question might have to be tackled again, but that's far away and I think with the infrastructure I have and can get with my allocated "solidary" budget, It'd really have to get to insane amounts to actually be an issue.
So I'd like to think it is indeed sustainable. I'm doing this to be sustainable. I want to build something people appreciate and use. I'm happy to spend lots on it!
Danke Leo,
Seems to work again! Very nice and fast onboarding. To be honest I signed up without a specific use case so have to figure out what I really want to do with the tool, but I'm impressed at the sleekness and speed. Eversthing sleek, unitary design, feels well developed and mature. Very impressive for a freshly launched product!
Small suggestions:
* maybe default-disable some of the modules people are less likely to use (e.g. finance). After doing the selection the onboarding-per-module tries to go through everything - maybe easier to start with 3-5 default and then invite people to add? It feels a bit overwhelming if you don't really know what you want yet. But that might all just be me.
* I noticed DE and ES are "du" but FR is "sie", not sure that was intended, I guess it's AI misinterpretating the prompt (or maybe the eternal DE-FR feud made the "vous" a conscious choice?). The German is also a little clunky but that's quite normal for translations..
* Some info on how to export would be great (e.g. MD notes)
* Maybe you want to make a slight narrative change, from "sign up for free" (which I read first as indicating that a payment requests comes later) to "sign up - it's free" and summarise a bit your philosophy as above to make clear that you intend to keep it that way - but maybe it's already there and I missed it :-)
Tolle Initiative und immer schön zu sehen wenn Leute versuchen die Welt besser zu machen!
Now a little more in-depth response. Thanks again for your feedback, I could relate to all points you mentioned.
- You are totally right. The "Try it for free" could very easily hint at cost down the line. I have changed it :)
- Cool that you understand so many languages :) And thanks for your suggestions. Looking into it right now. Translations is one of those things.. doing it by hand (not me, but someone with good language knowledge) is very time-intensive, but yea, AI can be hit or miss for it.
- I'm definitely thinking about starting with a leaner set of features. My only fear would be that people forget there is more to it. But that should be solvable via UI.
- Definitely going to add more info on export! That's a shortcoming right now which I aim to fix this week.
- Regarding philosophy, I'll think about it. I'm not great at marketing (and don't even want to look at it as "marketing"), so I'm always a little unsure what messages belong where. But good to hear, that a more detailed part on this could be good.
Hey gooba, danke danke danke!
I'll have a go through all your points tomorrow, greatly appreciated! Especially since when you work on something for so long, you just can't see things with fresh eyes anymore. So it really is great to get feedback like yours. So thanks again!
Hi, I tried to find more details of this initiative which sounds quite interesting (I'm living in Belgium) but it seems there is no website or summary info available - is the only way to learn more about the model a 4h lecture recording? Would you have any kind of package or info you can share on how this really works, wha the financing model is, etc?
We used to have a public website but Enernet is a technical and political solution for whole communities, companies and factories, not for single individuals.
Please send an email to Fiberhood at icloud dot com with the location of the proposed Enernet smart grid and we'll find out if there is enough interest in your neigborhood (around 10 participants) to set it up.
Enernet is a high tech solution for getting the best ,cheapest, sustainable energy, internet, mobile phone, transport (cars, busses, trucks) and housing in a neighborhood for people and companies.
If you wire up all buildings with fiber and cables to each other you can replace the national grid, its laws, taxes and always high prices with a solar energy system at cost price, saving several thousand euro's per house per year. Instead we put an abundance of solar cells up and install insulation, thermal and electrical storage tailored to the local situation. We fund it from the large savings we create. The participants will decide what will be installed, so every Enernet smart grid is different. Our Fiberhood coop has special electronics that make batteries a lot cheaper and safer and replace the National AC grid with a DC grid that has almost no transmission losses. We also distribute 25 Gbps internet and establish a small datacenter to heat water cheaply with the waste heat. The savings from all your utility bills pay for the installation and upkeep of the Enernet network.
As this all seems so straightforward I would be surprised if anything is anonymised or otherwise sanitised to preserve privacy or user's secrets.
reply