So this will mean a chat interface to communicate through hyperon? This will be much more advanced than normal DNNs like OpenAI ChatGPT? Any idea in which version of ASI:create this can be implemented?
Читать полностью…
Please skip price speculation and broader market discussions on this channel, thanks
Читать полностью…
You’re lucky he even took time to respond to you tbh
Читать полностью…
He did just try and explain.. I don't think you're trolling but being nice is free..
Читать полностью…
On a technical level, no. GraphRAG is one way to leverage KGs in LLMs but only sometimes the best way. If the KG is a Hyperon Atomspace, one can do better by translating the natural language in an LLM prompt into pattern-matching and unification queries against the Atomspace using semantic parsing, then executing these queries against the Atomspace and using the results to expand the prompt. However that is indeed "jargon" and to explain it in a nontechnical way would require elaborating a bunch of examples, which I don't have time for right now, it's midnight and I have a couple more hours of work to do before crashing. Good night all 😉
Читать полностью…
Isn’t that GraphRag as you explained earlier?
Читать полностью…
The only thing that I read from you is jargons after jargons. The only thing you did to differentiate GraphRag with Hyperon is to add more jargons.
Читать полностью…
If you cannot explain it, you don’t know it. The lack of a product from your team speaks volumes. And name calling? How classy.
Читать полностью…
I think 90% of our new community is unaware of your work, @bengoertzel. Please let your team create more blogs about this and also make the Twitter community aware of your work, specifically the vision you just explained.
Читать полностью…
Thank you for showing up here and these explanations. We need this kind of stuff in the roadmap or blogs with future plans for ASI. This is what everywant need to read beside this small telegram group.
Читать полностью…
Both sound exactly the same to me. The only difference is that Microsoft has a working product
Читать полностью…
Microsoft's GraphRAG simply takes a knowledge graph and transforms it into a big text that can be used to fine-tune a transformer or similar.
This is nothing like the Hyperon system which does probabilistic higher order predicate logic reasoning, probabilistic evolutionary program learning, algorithmic chemistry and a whole bunch of other AI algorithmics in a dynamic in-RAM knowledge graph spread across multiple machines in a decentralized network.
The commonality is that graph knowledge representations (though of very different sorts) are involved in both.
I think what ben says is that there are 2 approaches to ASI: full power but limited (o3), less power but unlimited (Hyperon). We can't compete with someone who only focus one power while our goal is power AND decentralization.
Читать полностью…
What you are doing sounds suspiciously like the one already released open source by Microsoft, the GraphRag
Читать полностью…
Are you depending on centralized big tech such as Meta to come up with a satisfactory OSS llm model before your Hyperon will even start working?
Читать полностью…
Sounds amazing! Looking forward to the more hyperon AGI-based roadmap looking into 2026 with a clear vision. Keep up the good work!
Читать полностью…
𝕵𝕰𝕽𝕸𝖀:
Thanks, Ben, for taking a little time to shed some light on the distinctions between approaches. The report as well!
I thought I might tag here with “how it works”. I’m assuming newcomers are like myself in that they use the TG channel search function to learn more about their favourite projects. Your comments give some broader insight beyond just the pinned messages.
Thanks everyone else for doing your best to stay respectful and collaborative!
Hey, I just wrote a long appreciation post that seems to have been automatically deleted (with the TG disintegration animation on the comment).
If moderators can revive it, that would be appreciated! 🙏
If not: general positive sentiments! 👌
Have a nice time running away from more difficult questions.
Читать полностью…
I will take it as a worthwhile TODO to create a simple explainer of some sort for Hyperon, which clarifies the core nature of the system for folks with no tech background. Thanks for the nudge.
Читать полностью…
Indeed the recently released roadmap focused on products to be released in the next 6-9 months, and on that time-frame our more advanced AGI R&D stuff will only play a supporting role...
But we are designing ASI:Create so that
-- initially it will focus on making it easy for AI devs to connect fine-tuned LLMs to knowledge graphs
-- however, we will then add more reasoning capability to the knowledge graphs themselves (which is straightforward for KGs in Hyperon Atomspace format)
-- As our scalable Hyperon codebase matures, the center of gravity will switch, and Create will become essentially a platform for devs to connect Hyperon instances to their applications, with LLMs serving to feed and guide the Hyperon instances...
So the ASI:Create roadmap, post its launch, will involve it gradually. morphing into being mostly a decentralized Hyperon front end...
There may end up being a chatbot comparable to a ChatGPT type system as well, but it will be implemented as just one among the various dapps running against ASI:Create ...
Sometime early next year we will release a roadmap looking ahead into 2026 which will then focus more on actual AGI rather than just near-term product plans ... both are important!
Dude... both sound exactly the same to my toy poodle as well, but so what?
Читать полностью…
You're exiting. You need to guide the marketing team with the blogs and roadmaps. This is so exciting and people need to read this.
Читать полностью…
My SNet/ Hyperon colleagues and I have been around the AI field quite a long time. We know the players. One of the two DeepMind technical founders worked for me and learned about AGI from me in the first place. I was at the OpenAI unconference back before they became ClosedAI. I personally know (in some cases well, in some cases casually) a pretty sizable fraction of the lead AI researchers at every Big Tech company in US and China (I lived in HK for 10 yrs). I've only been doing blockchain seriously since 2017 but I've been doing AI seriously since the late 1980s... I have organized the annual AGI research conference each year since 2008 ...
You can see what a tech company is working on R&D wise from the nature of their hires, among other things. OpenAI is a transformer-NN shop through and through. Meta is a DNN shop, transformers being a minor part of their story. DeepMind is way more heterogeneous, and they do have people doing some neural-symbolic stuff (as well as stuff in other directions like algorithmic information theory etc.), but not that much, and it's not the passion of their founders, Demis is really a bio-inspired guy....
No tech company has gone about systematically cherry-picking the best PhD students from profs doing neural-symbolic AI.... or evolutionary learning either for that matter.. not yet. They have done that for deep NNs pretty systematically...
The Chinese tech companies are pretty much copying the US ones, and the Chinese gov't (which controls the companies closely) has no desire to create systems with. more autonomy and initiative, they LIKE systems that stick close to the training data and are easy to control...
So no we can't know for sure what skunkworks projects are going on. But we know what major projects are going on.... It's pretty evident to anyone who is in the center of the modern AI field that there is no non-DNN-centered project of the size, sophistication or ambition of Hyperon going on in any Big Tech company or government agency.
Now, supposing we launch some super awesome open source neural-symbolic AGI system in late 2025 or mid 2026 or whatever ... how fast will Big Tech start to copy it? That's a quite different question. Probably very fast. On the other hand, at that point we will have no trouble getting immense resources for our own effort, and we will have a big head start in terms of having a dedicated team with their brains and hearts wrapped fully around our approach. And we will be fully decentralized, so ideally positioned to take advantage of a large global team crystallizing from universities in every country etc.
Interesting times are coming 😉
What does it mean by executing on the design?
Читать полностью…
Indeed not, we are executing on this design pretty rapidly right now: https://arxiv.org/abs/2310.18318
We are also giving out $1M+ USD in grants to enable more OSS devs to join the effort ... and likely an additional batch of grants early next year...
https://deepfunding.ai/all-rfps/
We don't need LLMs to make Hyperon work. However they can be a quite interesting accelerant.
You know when you have a meeting at 7am, but it is currently 6:50am and you are about 40 minutes away from your destination. You know that you won't be there until 7:30am (30 minutes late), but you look at your car's clock, and you convince yourself to feel a level of ease because it's still 6:50am, and you are not actually late until the clock says 7am, so you have 10 more minutes to not go into a panic, because you know you'll have enough panicking about being late in just 10 minutes (when you are literally late)...
I hope no similar psychological concept is being unintentionally applied at Snet in that since "there hasn't been any big tech announcing that they working on symbolic reasoning and evolutionary learning, it stands that they are not working on those models, so we are still ahead of the game."
My point is, how can you be sure they are not working on the thing you say they are not working on?
I mean, Microsoft kept the surface computer line pretty well under wraps and they did so with the hololens as well. So how can we be sure they are not keeping an atomspace with opencog hyperon equivalent under wraps until the right time to announce?
I find it unlikely that big tech is not exploring all model options
You seem to have a better solution, please share.
Читать полностью…
Is this a loaded question you are asking?
Читать полностью…
