alxfoster

Your annualized energy cost estimates are off by an order of magnitude. 1kwH @ $0.1 (Texas) is $2.40/day if 100% utilized 24x7, California is roughly twice that per my understanding.

Unfortunately I think you’re on to something here. I love ‘vibe coding’ in a deliberate directed controlled way but I consult with mostly non technical clients and what you describe is becoming more and more commonplace -specifically within non-technical executives towards those actual experts who try to explain the implications and realities and limitations of AI itself.

Honest question: how does this calculus change when the person waiving a foreign flag on a burned out vehicle is actually not a citizen and armed with a weapon.

Lets separate headlines from reality here: Yes this is an unnecessary provocation with loads of emotionally charged elements (and federalizing California's National Guard in this context is certainly concerning for multiple reasons -considering the scale of the protests and violence ) BUT there is no sign that Active Duty military personnel are being deployed to engage civilians (yet).

It would seem most likely that the Marines were called strictly to protect federal buildings, facilities and agents. The problem I see is the latter category. I am personally fine with National Guard being used to protect people and infrastructure when appropriate and when confined to federal facilities, and I'm even fine with the use of military to protect federal facilities... however, the second active duty military engages civilians 'on the streets' we have martial law and that's a whole new can o worms with explosive possibilities for escalation.

IMHO, Monero checks every box. Bitcoin is not as anonymous as most think. Monero may be a little more difficult to exchange but last I checked, most major exchanges outside of Coinbase still support it.

According to the CDC, NIH and other respected credible, mostly objective federal health research groups have all suggested up to 2 drinks per day for men and 1 drink for women is not only safe but also beneficial, citing that moderate drinking "reduced risk of heart attack, atherosclerosis, and certain types of strokes". Obviously this would not be the case for people prone to alcoholism or some other complications or contraindications. Sources: https://www.ncbi.nlm.nih.gov/pmc/articles/PMC6761695/

https://www.cdc.gov/alcohol/about-alcohol-use/moderate-alcoh...

Lets be honest here: this article seems as if it were drafted by Altman himself. It's incredibly biased and screams pro-Altman agenda. I would be very surprised if " 90% " of any company could agree on anything, including the removal of the CEO. What is clear is that there were massive conflicts of interest and that the board probably did their job in preserving the mission of the organization (they sure as heck are not operating as agents of MS). The naive fanboy blind support of management here should be concerning to any rational objective actor who understands fiduciary duty and the bigger picture.

Frankly, I think the pathwai.org taxonomy is far more useful in the long-term even if many of the individual attributes are speculative. Frankly, I'm a little surprised that Deep Mind neglected to cite pathwai.org's research. I'd highly recommend checking it out (though full disclosure, as the author I have my biases.) http://www.pathwai.org/index.html (Desktop only, best at high resolutions)

Quick question regarding the scalability and support of multiple vector databases under a single cloud service. Suppose an enterprise Saas product served multiple customers with each requiring a unique RAG vector knowledge-base for product and company info. Do any of these solutions allow for a large number (dozens or hundreds) of small distinct Knowledge bases? Do any offer easily integrated automated pipelines for documents to be parsed and ingested?

FYI, many of us are indeed running 65B. I’m running 65B at 4-bit and getting about 7.5 tokens per second. Granted, I have a beefy machine with 2x 3090s and Nvlink but certainly well within the realm of any small lab.

Been trying for months: https://news.ycombinator.com/item?id=34646147

Hi All, This is a work in progress but am at a point where feedback would be helpful and I can't think of a more appropriate audience. The overarching purpose of the following project and taxonomy is to both propose standardized ordinal levels of AGI and accurately frame, track and predict the state of specific Human-Equivalent (HE) AGI components, particularly for Human-Level Sentient (Conscious) AGI. While this is admittedly speculative and hypothetical to varying degrees, it is a starting point that might allow for more accurate expert-sourced extrapolation of timelines to specific types of HE AGI. Clearly much debate exists within the cognitive sciences as to what exactly constitutes intelligence and moreover, consciousness and it is not my intent to make judgements on existing theories or the risks, dangers and alignment challenges, however, it is clear to me HE AGI comes with clear ethical responsibilities and we should be just as concerned for the rights and ethical treatment of Self-aware HE AGI agents as we are for that of humans. In fact, this reciprocity of respect may have more pronounced long-term consequences than we can imagine.

While many developmental paths to HE AGI exist (eg. Brain Emulation, Seed, etc) It is my humble opinion that our most likely initial path will follow brain-inspired cognitive architectures that will certainly include the 5 key categories listed (Perception, Understanding, Cognition/Reasoning, Consciousness and Actuation). Ethical considerations aside, this is likely our best option to allow for alignment somewhat analogous to moral consideration between sophisticated human. Each of these distinct systems will of course converge to produce consciousness and each will incorporate multiple DNN/LSTM/MORL/etc. subsystems. Sidenote: current GPT LLM systems for the most part 'deepfake' reasoning with statistical inference but I estimate a shift towards continuous learning systems much closer in architecture to our own neo-cortex within 5 to 10 years.

Feedback and constructive criticism is welcomed on the discord I setup for this project here: https://discord.gg/zj3HNQUXsf