https://cerebralvalley.ai logo
#02-general-community-chat
Title
# 02-general-community-chat
u

윤희상

12/17/2023, 11:25 AM
Hi, I’m planning to move to San Francisco, starting from next February to bootstrap our AI healthcare startup with my co-founder. We are trying to increase human life expectancy with AGI. We’re currently pondering on locations — Where do you guys recommend? 1. Hayes Valley 2. Rincon Hill ~ SOMA We’re looking to live nearby peer AI startup founders and engineers. We also want to attend meetups from time to time.
👀 1
Or maybe other options as well — 3. Russian Hills ~ Nob hill 4. Pacific Heights ~ Japan town Feedback is welcomed.
m

Michael Wolfe

12/17/2023, 6:28 PM
Do you have an option to find a temporary place to live for a few weeks so that you can get more familiar with the city and how the neighborhoods differ? The city is small enough that I’d probably recommending finding a neighborhood that matches your lifestyle, culture, and hobbies vs. optimizing for work. For example if you love to get outdoors you’ll want to be in the northern part of the city or near the park.
👍 3
u

윤희상

12/18/2023, 3:04 AM
@Michael Wolfe Thanks for the advice. We don’t really have an option to look around for a few weeks first, we’re preferring to commit to 2~3 months of housing first (we’re moving back and forth because of VISA). Totally agree that each neighborhood has a very different vibe. Is there a sort of a hot spot neighborhood for peer AI startup founders and engineers? Or is the city small enough to not care about minor proximities?
m

Michael Wolfe

12/18/2023, 3:08 AM
Probably Hayes Valley, the Mission, or SOMA
👍 1
🙏 1
u

찬종나

12/18/2023, 6:41 AM
Hmm 2-3 month duration is very short. I would doubt you finding afffordable housing with that short term. You are likely have to pay 1000-1500$ more per month for that. Any reason why you want live in the city? I feel like if you expand your option to bay area you get better deal as long as you live close to Caltrain station
Regarding hot spot neighborhood, I guess https://agihouse.ai/ this is a good bet.
👀 1
+ Kinda seperate, but I am curious of your wording 'AGI'. What do you think is the current/future capability of LLM
u

윤희상

12/18/2023, 8:55 AM
@찬종나 at this point of browsing, $1000~1500 seems super affordable to me 😅 AGI house in hillsborough is a great option. I know they host some great hackathons, but dunno how we can actually live there. Perhaps do you know how?
u

찬종나

12/18/2023, 8:56 AM
Oh no I meant u'll pay 1000-1500$ on top of market rate. So should be around 3500-4000$ Don't quote on me tho lol
1
They have a form which u can apply
1
u

윤희상

12/18/2023, 9:00 AM
The price range of our current AirBnB candidates are offering at around $2000/month. + I thought the AGI house’s forms are just for the events/hackathons
As of LLMs and AGI, I mostly agree with the definition of superintelligence proposed by deepmind Betting on the scaling law for the foreseeable future.
u

찬종나

12/18/2023, 10:14 AM
I feel like current transformer architecure is already hitting a rock bottom. Just increasing pararmeter size is not providing sizeable return. (Diminishing rate of return). I heard rumors of proprietary models utilizing multiple mix of techniques on top of transformer model. That is why I am sketpcial of AGI as even SoTA model like GPT4 / Gemini are quite bad at simple intelligence task for math for an example.
That being said, my theory is we can still utilize specialized LLM for targeting specific domain for decent success rate in foreseeable future
u

윤희상

12/19/2023, 2:00 AM
@찬종나 I don’t think we can say that transformer architecture has hit rock bottom yet, since the SOTA is still GPT4, and we don’t know if OpenAI has hit an internal limit or not. Maybe they are adding different minor techniques to increase performance for GPT5, which seems like a deja vu of the RNN days before transformers. Certainly a possibility. Would like to hear more of your reasoning. The overall trajectory and trendline still seems promising to me.
u

찬종나

12/19/2023, 5:04 AM
What GPT is using under the hood is https://huggingface.co/blog/moe So simple transformer model proposed at Attention is everything is already very outdated & inadequate for training larger models. However slapping more parameter willl increase evaluation metrics. How much of that is real? Is also debatable. It's pretty easy for proprietary language model creators to add evaluation question and answer to the pretraining to massage evaluation accuracy Take a look at this https://www.interconnects.ai/p/evals-are-marketing
👍 1
That being said I think OpenAI should be closer to truth than public companies like G or M or A. Public companies need to pump stock price every quarter, thus more likely want to show some impressive stuff, albeit massage model to be just good at evaluation
k

Konrad Gnat

12/19/2023, 10:21 PM
for housing, brownstone.live is a good inexpensive option. The location right in SOMA on 5th and market is great.
👍 1