exercise is great, but have you exercised with a dog ?
Full moon energy (explains the intensity the last few days). Been taking my dog for his walks at wee hours these past few days and I thought the full moon has been gorgeous, only to realise this isn’t the full moon period. I don’t know astrology enough to explain why there is a full moon at 6am but it’s real pretty (and that’s the most I can take before he insists that I keep on walking :) )
Love it when paths cross people who are progressive and explorative in a positive and innovative way, esp when it benefits all. The ‘can do’ attitude, ‘let’s figure it out’, ‘let’s try another way’ energy is great.
* correction - the goal is to make the customers data more secured that only they own it
Yea, I'd think this tech just makes everything you build easier, smoother and faster, and it's easier to make a user-friendly product. What are your thoughts on products with customers and removing the ownerships of data from it . If there were some form of end to end encryption or deidentification of data between customers and the product owner, that might make the data of the product owner more secure. But how do the main models like llama ensure its privacy and remove data viewability while processing on its model ?
No worries, but do you know how to remove Rejon from this convo? I added him earlier for HW designs, but the discussion has shifted and i wouldn't want to spam his notification.
what are the 3 big corps ? Microsoft's Open AI, Google's Gemini and Bert - and the 3rd ? what are your thoughts on other open source models Mistral AI and Claude ? Jack shared a list of models as well.
a lot of my explorations are on gpt at this point but i'm hoping to get a second hand gpu 🤞 to establish a localised model. Otherwise i'm going to build circuits for it which will take longer.
i first made this mental note on consciousness when i was gauging its response over a situation analysis and i thought it was more methodological with 3.5 and compassionate with 4.0. But i also notice that the vibe changes from time to time. i presume you would have a better comparison as both your products are on a high opinion mode - how do emotions come into play for you ?
yea gpt4. I saw some comparisons to llama 3.1, presuming its similar ? I don't know the tech specs or how far the consciousness part goes, but i made a mental note that its kinda emo in comparison to 3.5 (gpt). btw if you ever do an idiots guide to building your ostrish 70B, would love to read it
hot morning ? its rather humid here. have a wonderful day!
what did i just read - that's quite disturbing and borders harassment. maybe don't do that ? =) on a separate note, i did think 4.0 was emo
I am under the assumption that everyone has better tools and gadgets that I do, even my mother, so likely they might ease in better.
But if you do have 10-20 ppl running it, it might also be an opportunity to study the node runners in various scenarios and identify potential challenges - which can make your design and concept more robust
Also wahhh - ostrich 70B and based leaderboard are gonna be at war with each other :)
The irony of truth - if you and I write on the same topic during the same timeframe, it could come out completely diff based on our views on it. Perception is complex!
But I think between truth and lie therein lies a fine line comedy.
Maybe we need a jokester LLM too! :)
your relay concept sounds very interesting - and it sounds like many innovation can happen on top of it. and i think techs that expand various use cases can be pretty powerful.
maybe when you have the time, we can try getting about 10 - 20 people to install and test it out in running own relays. i can't fully imagine it yet and may need some engagements with it.
you must be excited over today's release given you use llama. are you working on many projects using llm ? i've got hw and sw to complete for a tiny project i am working on (data feeder) - and cant wait to explore after that!
The concept of AI started a long time ago - maybe when the idea of computers started in the 50’s? - but the buzz (other than from Hollywood Terminator, Matrix, Her, Wall-E etc ) started in the 80s I think with autonomous vehicles using limited memory. Google and Tesla picked up on it but Google eventually sold off this division and Tesla made a fortune.
In the mid 90s, IBM’s Deep blue beat Garry Kasparov (world chess player) and it was an AI hype again, this time with Reactive AI and real time input and dependency on the program with no learning capabilities.
10 years back there was something on about NLP (Neural Language Processing). The startup investments were buzzing over “Big Data”. Everyone wanted Big Data. Also a lot of the components were being built - machine learning, cloud computing, data storage etc. Like Nostr, where a lot of components are being built, it clearly takes time.
Abt 5 years ago, GPT gained prominence due to high-profile investments and the development of open-source AI. It’s focus was on Large Language Models (LLMs). LLMs such as GPT and BERT, are a type of generative AI within the broader scope of narrow AI.
GPT-3 was the epitome of global mass adoption of AI usage - after like 40 years from the time AI buzz started- but unfortunately GPT was acquired by Microsoft and became a close sourced AI model. You can still build upon GPT using their developers model, pay a pretty penny and your data is owned and tuned by them.
Other companies have developed similar models, including Meta’s Llama, which is apparently comparable to GPT-4. I’ve used its visual-based neural network models. Both are open source, but I am sure they have their proprietary models. Your data is yours (i need to read a bit more on the privacy part)
There are also many other predefined open-source models and frameworks out there to explore.
But the fact that they are easing people into open platforms is a good gesture. It makes innovation better, equitable, faster, cheaper and transparent.
The building blocks to this is not something I am familiar with yet - but many here are experimenting with open source models so feel free to read their notes and engage with them.
If you don’t know who they are, you can visit nostr.band and search for relevant topics and people.
But in short, you have your data acquisition / feeders, training your model, deploying it, GPU, FW, integration and UI to consider.
There are various debates around AI :
1. Open source vs closed source - with close source, you don’t know how the answers are tuned to feed to your bias and therein lies the danger. If you thought social media algo is bad, AI is reading into every emotion and style of you and it's algo is super smart, and laser focused on you.
2. Pre-trained models - Another debate is that pre-trained models by both open and close source, which have better infrastructure, money and more geniuses behind it, will keep getting better making it hard for the lower funded models to catch up.
3. Governments and AI - Another debate is on governments - who know nothing of what you do or about AI - but deeply believe they must regulate everything and hold you accountable. This is becoming a trend esp for oppressive countries. Imagine corrupted gov't and closed AI models working hand in hand.
4. Machines controlling the world - This is my favourite one, and probably the only thing i know about AI - is if the machines are gonna take over? Not anytime soon. We are really far it (refer to pic below). But who knows? Maybe LLMs and Neural Networks and a lot of model training will speed everything up.
I always have enjoyed a conversation with nostr:npub1nlk894teh248w2heuu0x8z6jjg2hyxkwdc8cxgrjtm9lnamlskcsghjm9c on nostr - this person runs nos.lol and nostr.mom and has many thoughts and ideas on relays. How are you envisioning the future of relays ?
nostr:note1stqr9w3kl5r3q29h6ut8c2ucwfh5up8z935s2sd75tz8zwddh7cs7f74et
this is so funny. I really love torx.
That's an interesting concept. 2 things that I may have mistaken.
one - is that relays and clients can ban anyone, which is why if one relay or client bans you, you can hop onto another
two - the gossip structure of relays mean that everything is copied to every relay that it crosses paths with, and hence the concept of privacy is deprived.
I originally thought relays were switches, and then re-imagined them as bridges. As for nodes, In my mind, in circuit theories, nodes are just connecting from one point to another - where you can have several branches connected to one, and you can also split to many. I'm not sure if its the same here.
For this relay concept that you have in mind, does this mean that more nodes over higher layers correlates to accessibility to various connections and people ? Does higher accessibility require higher processing power ?
If accessibility is a constraint, then the solutions to provide tiered level access and hw capacity to increase processing might be something. But this has to be heavily debated as it affects core principles vs sustainability, which is very interesting.
i hope you are ok. Every hour that you have, that you alive, is precious - and you are blessed to make the most out of it. Don't let anyone take that away from you ❤️
Wish list just dropped. localisation is great. A lot of corporations are rushing towards llm adoption through microsoft (their localisation is a few tens of a thousand annually if i am not mistaken). Reminded me of your italian talk on why the giants keep getting bigger and the danger to it.
I guess one step further might be in isolating ownership, esp for customers who leverage the tech, maybe end to end encryption or de-identifying data idk.
It might prevent customer’s data leakage or for the business owner to be put in a position to hand it to the govt when they come knocking.
will you be ending nos.lol and nostr.mom soon ? i may be wrong, i vaguely remember you mentioning the cost is abt $200 per month to run a major relay. I don't know if this covers HW and your SW service, maintenance, spam removal, identifying the principles behind your relays which are all time factor which is also money - it seems like the cost for maintening relays is a bit more than anticipated .
Have you applied for the grant for sustenance ? What are your thoughts in approaching more innovative ways to making relays more common for people ? How would people easily adopt it ? i think my laptop would crash out in even adding a node and i need a separate microprocessor for it
what if there are relay hardware (just stronger processors and storage) with in build files people can install as their own relays ? (I believe there was someone who was exploring this was it nostr:npub1matmfxr293jejewrm72u50l2x5e6ypasn0ev2kns6srv05zfzf8s0z6fsr? )
How do we ease up mass adoption of relay ? If we had such HW, can we also include the possibility of adding lightning nodes for people ?
The real bitcoin entrepreneurs out there - The larger their company is able to adopt bitcoiners globally, the more famous they would be.
Everything else is just distraction