dougSF70

11 ms·

91.
by dougSF70 3y ago
They might also be most motivated by any incentives offered by the city to bring people and shops back to market street/ union square. Part of the perception issue is the ratio of homeless to non-homeless people. Tipping the balance b
92.
by dougSF70 3y ago
The lenders need to revalue the properties so as to lower rents. This could bring an influx of independent stores in. However, downgrading of commercial real estate does have broader ramifications
93.
by dougSF70 3y ago
This is what I don't understand, the capital cost of a machine to replace humans only works if the productivity multiplier > amortized capital cost / annual labor cost. However there must be a limit to the multiplier due to ma
94.
by dougSF70 3y ago
Funny story. I worked at a firm that was commissioned to do some market research on Spore before its launch. We were given some videos of game play elements to show gamers and get their response to the game ahead of launch. We hosted the vi
95.
by dougSF70 3y ago
This is exactly what some VC firms do. They are interested in generating returns for their investors, if the company trades to zero after an IPO they will have already generated a significant return.
96.
by dougSF70 3y ago
Also, Datacenter GPUs only have passive cooling, presumably to allow for more Cuda cores. I get that what I have is older tech but for the cost of one H100 GPU I can have 60 of these servers (600 GPUs) plus some change to pay for CoLo fee
97.
by dougSF70 3y ago
I bought these from a Chinese internet firm. They dismantled their data center and I am buying it.
98.
by dougSF70 3y ago
I have just bought 4 servers with 10 GPU (Titan X Pascal) and 384 GB of RAM for the price of a macbook. I found a solar powered CoLo facility to put them in. The economics of owning vs renting from AWS are worth it. My server set up is the
99.
by dougSF70 3y ago
Self driving taxis are the domain of cash-rich and time-rich people. Like an uber it just takes 30% longer to get there on account of the law-abiding nature of the "driver"
100.
by dougSF70 3y ago
If you are on strava , you can see my training log.
101.
by dougSF70 3y ago
It can be done with 16 weeks of training, 3 runs a week 10k, 15k, 20-25k. Each run you just need to train running the last 5k at a faster pace.
102.
by dougSF70 3y ago
At the age of 52 I ran my first marathon, not an organized race, I just picked a day, trained and ran it with a friend (an Ultra runner) who paced me round. I will never be fast but the sense of accomplishment of even a 10k or a 5k makes ru
103.
by dougSF70 3y ago
Totally agree, substance over style.
104.
by dougSF70 3y ago
I have just purchased 5% of a data center's Deep Learning servers. 10 GPU / 384 GB of RAM, 2 x Xeon 10-Core processors. I am putting them in a solar powered data center and plan to lease them out at $0.2 per GPU hour.
105.
by dougSF70 3y ago
This reads as: Scientist discovers powerful genie in a bottle. Scientist releases powerful genie from bottle. Scientist now regrets releasing genie from the bottle.
106.
by dougSF70 3y ago
A classic False Economy at play.
107.
by dougSF70 4y ago
If you go into Firefox Focus settings and set hype_tolerance to 11, then you will see it.
108.
by dougSF70 4y ago
Perhaps it is more the looseness of the English language that could lead to less performant LLM models, rather than a critique of any particular languages vocabulary. The Japanese example was just what sparked the thought.
109.
Ask HN: Would GPT-n produce better model if trained on a structured language?
2 points by dougSF70 4y ago|3 comments
110.
by dougSF70 4y ago
Now I know where to build my LLM focussed data-center
111.
by dougSF70 4y ago
You must be thinking of PaaS, priest-as-a-service.
112.
by dougSF70 4y ago
SaaS, sacrifice-as-a-service
113.
by dougSF70 4y ago
You can buy the Costco store inside the Costco store and then take it to another Costco store and get a refund...great customer service
114.
by dougSF70 4y ago
Perhaps this is a concerted, organized legal strategy that Georgia, NYC and the Feds have organized.
115.
by dougSF70 4y ago
They bid until they are red in the face.
116.
by dougSF70 4y ago
One other thought I had, was did anyone fact-check the training corpus. It is probable it contained factual errors, which would mean that the trained model would propagate errors some how.
117.
by dougSF70 4y ago
Since the LLMs are probabalistic is it possible for the inference engines to highlight lower probability content. That might at least make them more "human" rather than being confident and inaccurate.
118.
by dougSF70 4y ago
The twitter thread they published appears to contain supporting evidence. Or perhaps there was enough smoke to justify the short position which they covered at market open for a nice return.
119.
by dougSF70 4y ago
I have often thought this is the business model of most payment apps.
120.
by dougSF70 4y ago
Block stock price down 20% at market open...Hindenburg have generated a significant return already.

More ›