
Founder: https://xitter.cf/t.co/Oimx2qEGLj, https://xitter.cf/t.co/zMZiD2uLfP & more ; ) @uncensored_ai
Austin, Texas
Joined March 2019
- Tweets2.3K
- Following105
- Followers461
- Likes10.2K
Pinned Tweet
I last tweeted in 2023
Since then:
- built top 200 ai sites in word by traffic: uncensored.com
@uncensored_ai
- nurtured & disciplined a group of animals into being productive members of +ve sum society
...
sraj is a high agency lane opener typa guy
it's been a week since i joined uncensored.
the initial days, 1-2, i had a lot of difficulty setting up the codebase and raising my first pr and all that,I was pulling like 10-12 hours , but i do remember interacting with my teammates and asking them, "what's your work schedule?" and their response was all the time, was what work schedule they work all the time !!! no holidays, nothing like, the team is so good, everyone is so supportive, so young !!!!, talent density is crazzzy
it really pushes my ambition over the top. i've been sitting in the same chair for the last 10 hours, going to sleep now, but for some reason, i am pumped uppp.
bruhhhh, let's see how my time here goes and what value i can provide. i don't think i should tag them, but let's just do it @uncensored_ai good night
Blake retweeted
Apple announced LLM-powered keyboard for iMessage
It learns how you type
Major benefit to iMessage interfaces
Major benefit to users of obscure vocabulary domains like crypto
Expect trending & Overton-window-leading vocabulary to catch hold quicker
xitter.cf/drjimfan/status/166577…
From what I'm seeing so far:
@Google's text-bison:001 is a real contender to gpt-4
~10x cheaper
~2.5x faster
Similar max total token context window
google: ~9.2K
openai: ~8.2k
In tests with identical ~4K token prompts:
- Google text-bison took ~4.5 seconds
- OpenAI's GPT-4 took ~14.5 seconds
Yet, the outputs (commands) were identical
Response structures are alike:
-> google's: predictions[0].content
-> openai's data.choices[0].message.content
Google downsides thus far:
- Poor DevEX - high friction onboarding, including bugs in their dynamically generated GUI code
- Worse api monitoring options
Note:
I'm not referring to gpt-4 32k here (which has an absurd price per token at minimum $0.06 per 1K tokens, ~16x that of text-bison)
Blake retweeted
Replying to @danielgross
FYI
Can use kmeans clustering and/or sampling methods to generate the most salient "suggested questions" to ask about the pdf
basic example of this:








