Basement AGI is very likely. Barriers to entry very low (for those with the right skills/knowledge/insights). Most of the assumptions held by the industry today will be obsolete in just a few months. Large compute, big data, scale... these are what people like Roon assume will be
you can't trust elon, you can't trust sam, nor zuck, nor dario, nor satya, nor sundar, nor demis, nor sergey, nor larry. you can't trust any of them with the fate of humanity.
This is too good.
The authors behind this idiotic MAGIC paper on arxiv have NO AI background. I don't even think they know how to train a model let alone build one from scratch.
Incredible how this is a recurring theme amongst doomers/safetyists...
don't underestimate a small highly motivated team in the AI race. google has all the talent, all the compute, all the capital.... and yet after an entire year they can't even come up with a decent model to compete against OAI and open source AI.
yes we are resource constrained,
So Mistral prohibits you from using their models to train or improve other models or compete against them........... I thought they were fully open......
🚨 ATTENTION 🚨
The following companies in their TOS prohibit you from using OUTPUTS of their models in the TRAINING or IMPROVING other models unless it's to fine-tune their models:
Anthropic*, Meta Llama, Mistral, OpenAI
Open Weights, Open Outputs!
* Anthropic hasn't yet
To me the only thing worth building immediately is AGI.
Want to solve longevity? Build AGI first.
Want to build scalable functional robots? Build AGI first.
Want to invent Programmable Matter? Build AGI first.
Want to make fusion energy work in a reasonable time frame? BUILD AGI
Ok this is awesome, Reinforced Self Training, new RL finetuning method. 1 more step towards fully autonomous machines and the beginning of the end of manual finetuning (1 yr tops)
if anyone's looking to train Mixtral (full finetune) using axolotl by
@winglian
, i suggest you add the following configs to prevent train loss instabilities
since basement AGI is likely, i think we need a new type of VC fund/capital allocator with a thesis focused on supporting and accelerating the AI cambrian explosion. Allocating capital and resources to every promising path in the AGI tech tree. It will pay off.
should we build
@ylecun
@elonmusk
@jeremymstamper
@bneiluj
@how_many_roads_
well you don't really need to publish a paper if you can show it actually working in the real world. does spaceX have to publish a paper on reusable rockets for us to believe that it actually freaking works
if anyone solves ARC-AGI before me, and you don't feel like giving it up for $500K, reach out to me and you'll get leading position in our company with a starting comp pkg of $10M
A letter to OpenAI, Sam Altman and AI safetyists as a whole:
F*ck you for thinking that humanity cannot be trusted with open source SOTA AI, for manipulating the world to fear openness and transparency, and for elevating yourselves as the only rightful stewards of this
AGI for all. Robots for all. Abundant and dirt cheap (practically free and high quality) food, healthcare, housing, education..... (through abundance, not reallocation). No taxes, limited government. Maybe UBI but wouldnt even be needed with abundance. Do something you're
I've replicated and releasing Unnatural CodeLlama 34B.
Python-34B finetuned on Unnatural Instructions with QLoRA.
Humaneval and MBPP Evals underway with the help of
@abacaj
🚀
one thing led to another and i ended up frankensteining Mixtral into a 2.87T parameter MoE. Did a few tricks to make sure it's very memory efficient and it actually runs lol. It's 5x slower at inference than 'vanilla' mixtral. Still troubleshooting a few things and then I'll see
Superintelligence is within reach.
Building safe superintelligence (SSI) is the most important technical problem of our time.
We've started the world’s first straight-shot SSI lab, with one goal and one product: a safe superintelligence.
It’s called Safe Superintelligence
#ChatDirector
is a research prototype that brings 3D avatars and automatic layout transitions to your 2D laptop screen, transforming online meetings to be more immersive and dynamic. Check it out →
i'm assembling a team. dm your creds if you want to join the race to build generalist intelligence (agi) and tackle some of the key challenges on the ai tech tree
Elon is too diplomatic to start a beef with China so I'll do it for him.
We can't have China keep stealing IP from the West without any repercussions.
China is externalizing R&D costs and focusing all its capital in manufacturing, no wonder it's kicking our asses on the latter
$16K today.
$2K within the next years.
Update your priors.
All physical labour will be automated within the next 5 years. And we haven't even reached AGI yet.
Unitree Introducing | Unitree G1 Humanoid Agent | AI Avatar
Price from $16K 🤩
Unlock unlimited sports potential(Extra large joint movement angle, 23~34 joints)
Force control of dexterous hands, manipulation of all things
Imitation & reinforcement learning driven
#Unitree
#AI
a lot of ppl think that Google or MSFT or OAI have/will dominate AI in perpetuity. i bet none of them will, even if they boil the ocean. it's an inherent flaw in their business model and tech stack. the field is wide open, and open source will be a fundamental pillar of future AI
The future will consist of
- a small number of open source inference code,
- free pre-trained base models, and
- crowd-sourced fine-tuned models, on top of which customized (possibly closed source) products will be built.
Frankensteining models is a phenomenon that only exists thanks to a bunch of scrappy GPU poor open sorcerers to cut costs on crafting new larger and smaller models from existing pretrains and finetunes, rather than training from scratch. Truly a form of digital alchemy.
@nisten
some ppl look at AI and see a website or app builder or even a girlfriend (😬). others fear of being replaced.
while a tiny minority of us understand that AI is an empire builder that will usher in a modern renaissance.
I just asked one of my models to write me some python code, it ended up writing code to instantiate another version of itself from HF and asked it(s offspring? lol) to solve the problem 🤦
not much needed to replicate this OAI 'experience'. if anyone wants to join efforts to put all of what open source has developed into a cohesive platform such as this, ping me. it's within reach.
For those that say 'physical labour is a long way out', update your priors. Both digital and physical will undergo a fast transformation, especially as digital intelligences will help us build better physical systems.
Introducing 𝐌𝐨𝐛𝐢𝐥𝐞 𝐀𝐋𝐎𝐇𝐀🏄 -- Hardware!
A low-cost, open-source, mobile manipulator.
One of the most high-effort projects in my past 5yrs! Not possible without co-lead
@zipengfu
and
@chelseabfinn
.
At the end, what's better than cooking yourself a meal with the 🤖🧑🍳
They've gone full mask-off: 𝐝𝐨 𝐧𝐨𝐭 𝐞𝐯𝐞𝐫 trust
@OpenAI
or its products (ChatGPT etc). There is only one reason for appointing an
@NSAGov
Director to your board. This is a willful, calculated betrayal of the rights of every person on Earth. You have been warned.
OAI: "Trust me bro, we are 'capped' profit". Still pretty skeptical lol.
Alright then, what's the cap? When you reach the cap, what's stopping you from upping the cap? Your track record has been to compete against your own customers. Stop the cap.
If you are skeptical of
@OpenAI
’s non-profit / capped profit structure, please consider reading this page which goes into the detail.
I see so many people confused about how the structure works, share it with a skeptic!