bargainman
-
Posts
944 -
Joined
-
Last visited
Content Type
Profiles
Forums
Events
Posts posted by bargainman
-
-
Interesting interview from Karpathy (one of the AI luminaries) on what makes Elon unique
-
Hmm apparently Hussman is a legendary investor. Anyone know what his returns have been?
-
On 3/6/2024 at 1:54 AM, mattee2264 said:
Isn't it a little worrying for Mag7 investors that a start-up founded only a few years ago can release LLMs that compare favourably to the ones put out by OpenAI/Microsoft and Google? Reinforces the point that in AI no one really has a moat at this point. Mag7's ability to stockpile chips gives them an edge and they can outspend everyone on R&D. But in the internet age it was Google who ended up with the dominant search engine not AOL/Netscape/Microsoft.
Any company with a good idea/product will have little problem attracting funding and users aren't locked in to a specific LLM at this point so will switch if something better comes along.
And so far it is LLMs that are generating all the buzz and drawing in all the users and seeming to have the most practical use as a lot of people are using LLMs to write emails, assignments, research papers, marketing copy etc.
Made me LOL. Sorry but this is the nature of tech, always has been. Gates always said he was more worried about a couple of folks in a garage cooking something up that would turn the entire industry upside down. These days a LOT of stuff is out in the open. Take a look at huggingface which hosts community and models of many many open source LLMs. This is academia driven sort of. they always were driven to publish or perish and release their stuff to get reviewed.
These days though the limits are starting to show with training. It's crazy how expensive it is to train an LLM, how many GPUs and how much energy it takes. it's really only very large companies with huge budgets, GPUs and data that can afford it. Plus the data annotation process is difficult and error prone, nevermind ethically challenging.
Still, once models are out, a lot of them are open source, so you can finetune for less money, and just try prompt engineering to get what you want/need.
It's still day one.. early early days which is shocking but true.
-
On 3/4/2024 at 11:10 AM, rogermunibond said:
Isn't prompt engineering like Google-fu?
It's more than that. I mean there's user level stuff but then there are things like https://www.langchain.com/langchain that create components that get chained together for different flows. It's usually not just a single statement. When you message Chat GPT for example it doesn't just take what you wrote and give it to the LLM. It goes through layers and tries to prevent attacks and uses what are called "system" level prompts to wrap your user level prompt. Plus there's no memory so it actually sends your entire conversation through the LLM every time. So there's a lot too it.
-
Small positions in SCHD and DGRO. Getting older I guess lol
-
On 9/10/2023 at 8:03 PM, UK said:
The book on Jobs is also very good, for me the best, related to Big tech.
You have to take him for what he's like which is that he's not knowledgeable enough about the industry to really see things as they were or are. He got attached to Jobs, and has/had the reputation of being 'soft' on his subject. Apple didn't like his biography of Jobs since they didn't think it was an accurate representation. In fact they 'endorsed' a different one.
https://www.businessinsider.com/apple-execs-criticise-steve-jobs-biographies-2015-10
That's not to say he's a bad writer or biographer, just that he's biased and flawed like anyone.
-
On 3/14/2024 at 7:06 AM, Gregmal said:
Totally. I remember making a rather irresponsible earnings wager on Apple in maybe 2012 or so and being told "thats silly, its $400B company...how much bigger can it get?"...these things are silly talking points.
It's amazing. I remember after 2008 and 2009 when BRK was already one of the largest most well known companies. Since then it just kept going up up up.
-
thank goodness! 6% is ridiculous!
-
Well there's a whole new field of so called "Prompt Engineering", which talks about how to properly ask questions of the LLMs to get good answers. if you are serious about using them you probably want to check that out.
This is a free short course for developers but might be useful for others
https://www.deeplearning.ai/short-courses/chatgpt-prompt-engineering-for-developers/
There are likely more suitable ones for the chat interfaces.
-
10 hours ago, tnp20 said:
It is so early in this game ...this is a 30+ year revolution unfolding ....innovation happening so rapidly that I can not keep up on a daily basis - I am signed up for like 20 AI newsletters....
Today - microsoft came out with a 1 bit model (normal is 16 or 32 bit)....lower the level of quantization , faster and more efficient it is with what was thought to be loss in accuracy but thats not the case
YOu guys have seen Openai's SORA mode
www.openai.com/sora
now you can take a picture and turn it into a talking picture.
The media industry is going to be revolutionized ....
I assume you have watched the talks by Ray Kurzweil and Tony Seba?
-
With regards to the comments around unskilled immigrants, good luck finding nonimmigrants in the tiktok generation willing to do the menial difficult jobs like the migrant workers on farms, the dishes and floors in restaurants etc.
-
8 hours ago, Castanza said:
I rarely eat at MCD's but the wife and I were road tripping and stopped for a "quick and cheap meal."
Small fry - $3.69
Large fry - $4.95
Quarter Pounder meal - $13.69
McChicken - $3.99
McDouble - $3.19
4-Piece Nugget - $3.69
We kept driving...absurd prices for borderline garbage food. Can go to a local sit down burger joint for those prices.
if you get the McDonald's app, there's always a daily 2 for 1. 2 double cheeseburgers or 6pc chicken nuggets for $3.39. plus you earn points for the occasional free ice cream cone
-
14 hours ago, Sweet said:
I don't like the treasury ETFs for the following reason - please correct me if I am wrong.
If interest rates rise, the ETF will drop and if you want to exit the position you have to sell the ETF and take the loss.
However when you buy an actual treasury you, and the interest rates rise, you can just wait for redemption at par to exit (if it is a short term bond).
So if it is within your means to purchase a treasury why would you buy the ETF?
Well it depends sgov is 0-3 months so it really won't go down much. SHV is short term but slightly longer than that.
-
I get good rates from the cash in Interactive Brokers. But for my other not so fortunate brokerage accounts I use some combination of SGOV, SHV, and MINT. that said if it's in a taxable account you probably want to wait till it goes ex-div, otherwise you will be converting some part of your capital into taxable income.
-
On 2/28/2024 at 3:11 AM, mattee2264 said:
Something I have seen doing the rounds on Fin Twit is the idea that the release of Chat GPT is equivalent to the launch of the first web browser in the early 90s which means that we are only just getting started.
Its day one, and the investment and innovation going into this is astounding... As Is the pace
-
Has anyone tried this with Buffett's latest letter yet?
https://www.adobe.com/acrobat/generative-ai-pdf.html
-
-
On 2/9/2024 at 1:49 PM, ValueArb said:
Tucker just let him spew his fabrications and historical lies without any pushback, like any good journalist would do;)
What parts do you think were fabrications/lies?
-
At some stage LLMs will make their way towards on device inference. Some of the new google models are device friendly.
I'm sure Apple has designs on this as well with their neural engines and coreML, as well as the shared memory architecture.
-
On 2/23/2024 at 1:46 PM, Parsad said:
https://www.yahoo.com/entertainment/tyler-perry-puts-800m-studio-000722544.html
Just the beginning! Lots of jobs will be obsolete over the next 30 years. Cheers!
Indeed, it's always "day one". No one asks about how much energy this is all consuming though, and where the new content will come from once everything is generated.
-
On 12/30/2023 at 5:27 PM, mattee2264 said:
So generative AI doesn't seem to understand the concept of copyright. And copyright infringement could become a major issue with a lot of lawsuits. NYT has already sued OpenAI/Microsoft. Disney seems also to be weighing up filing a lawsuit. Seems like it could be something that could slow the progress until it gets sorted out and perhaps pour some cold water on the Ai enthusiasm in 2024. Any thoughts?
Some companies are focused on this
https://helpx.adobe.com/firefly/faq.html#training-data
-
LeCun and Karpathy are luminaries in the AI field. Yes LLMs are pretty impressive. Multimodal models are too. Lots of big challenges ahead but moving at breakneck speed.
-
I never really understood the rationale for the railroad purchase. The hailing of it as a high capital requirement business as if that was a huge positive. Now years later bemoaning that high capital requirement. I mean the higher that requirement, the higher the risk of a low ROE/ROIC no? I guess he did mention that he got it for a good price. Has anyone done an analysis of that part of the business since purchase? I wonder if he'd have made better by just buying back shares?
-
Not sure where I'm going but I heard there's a dude living on cruise ships 50 weeks out of the year! Apparently costs about 80K years which isn't so bad really...
Energy for AI
in General Discussion
Posted
Already happening to some degree with coding assistants like GitHub copilot and others. MS Office copilot possibly as well..
Question as it is with LLMs is how do you evaluate? testing is generally still done with human trials which are slow and very expensive. It will be a while before it moves to pure simulation if ever.
I wonder about this. If what Tony Seba projects comes to pass, I wonder if oil will completely collapse along with the dairy industry among others.