Opens profile photo
Follow
Click to Follow scottleibrand
Scott Leibrand
@scottleibrand
Built Open Artificial Pancreas (
@OpenAPS
) with
@DanaMLewis
. Day job
@Netskope
. DMs are open or email Scott@OpenAPS.org.
Seattle areaOpenAPS.orgJoined May 2011
519 Following
3,794 Followers

Scott Leibrand’s Tweets

Scott Leibrand Retweeted
Ethan Mollick
@emollick
So the lesson is anyone doing hiring needs to lean into doing structured interviews (they have to be done correctly, unstructured interviews where you just talk to people and ask random questions backfire)
Quote Tweet
Ethan Mollick
@emollick
The best way to assess talent in hiring is structured interviews. To do it: 1) Identify skills & traits you need 2) Ask about past behavior around those 3) Follow-up with probing questions 4) Ask everyone the same thing Here's some 💯questions to consider: saintpeters.edu/wp-content/upl
Show this thread
Image
Image
Image
Image
1
28
Show this thread
Scott Leibrand
@scottleibrand
As long as large context windows remain orders of magnitude more expensive than storing “memories” alongside lookup embeddings, it seems likely that at each step toward (and hopefully beyond) AGI that amplified systems will remain corrigible and debuggable. Thoughts? Pointers?
1
Show this thread
Scott Leibrand
@scottleibrand
Using approaches like
@AnthropicAI
did with their constitutional AI efforts, inspect such intermediate outputs for alignment, and correct any misalignment in the engineered systems before using them for any distillation of black-box building block models for further efforts.
1
1
Show this thread
Scott Leibrand
@scottleibrand
It appears that the easiest way to build amplified systems using multiple LLMs passes with
@LangChainAI
will result in “thoughts” and “memories” that are expressed in English and fully inspectable, both by humans and by other systems built with LLMs.
1
1
Show this thread
Scott Leibrand
@scottleibrand
Then the output from these amplified systems are used to train distilled models that more quickly approximate their output. Then those distilled models become building blocks for building more capable engineered systems, etc.
1
1
Show this thread
Scott Leibrand
@scottleibrand
Is anyone working on something like “corrigible amplification and black-box distillation” as an approach to AI alignment? First, simpler memoryless LLMs like GPT-3 are combined into debuggable systems with amplified capabilities (e.g. with
@LangChainAI
).
1
1
Show this thread
Scott Leibrand Retweeted
Becca Dzombak, Ph.D.
@bdzombak
My work for
@bioGraphic
: western Washington's Skagit Valley has a snow goose problem, largely of its own making. Read the 🧵 to find out how the community got into this situation, and how they're trying to balance flocks and farmers.
biographic.com
Washington’s Runaway Snow Geese - bioGraphic
Mae West said too much of a good thing is wonderful. But she’d never seen the beautiful, marauding snow geese that swoop in each fall to take over Washington State’s Skagit Valley.
2
14
Show this thread
Scott Leibrand Retweeted
John Semmelhack
@JohnSemmelhack
Another reason to love e-bikes: If you happen to have a family member who's a World Tour pro and former hour record holder, you can still keep up, even at 70+ years old or 7+ months pregnant.
Quote Tweet
Alex Dowsett
@alexdowsett
I’m a massive fan of E-bikes because of the freedom it gives me to ride with anyone. My Dad & Chanel in her 3rd trimester of pregnancy.. both enjoyed rides together as equals.
Show this thread
Image
1
8
Scott Leibrand Retweeted
Brendan Pierpont
@brendanpierpont
Gloriously wonkful thread. A lot riding on getting the H2 PTC emissions methodology right.
Quote Tweet
Nathan Iyer
@NiyerClimate
If we get this hydrogen thing right we will solve a huge chunk of power sector decarb. Y'all might not see it, but we have a windfall that could solve a major market issue. I'm gonna walk through why the H2 PTC is hot. Not even gonna try to be relatable here, wonk level max.
Show this thread
7
Scott Leibrand Retweeted
Casey Handmer, PhD
@CJHandmer
The general perception of outsiders is that solar deployment is, nay must, be tapering very soon. But if you back test that assumption against investment and construction along the production chain, you find that solar is just barely getting started. Hang on!
Quote Tweet
John Raymond Hanger
@johnrhanger
Good morning with good news: Solar capacity in US will rise by 85 GWs in 2023-2024. Wow! In each year 32 GWs of utility scale solar will begin operating. Small solar will add 9 GWs in 2023 and 12 GWs in 2024! 2022 solar capacity rises about 67% by 2023-24. eia.gov/outlooks/steo/
Show this thread
5
56
Scott Leibrand Retweeted
Tom Wenseleers
@TWenseleers
Whatever immunity gap there was - for influenza, it seems that in both the UK & the US it just caused the influenza wave to come earlier than usual, but the cumulative total influenza hospitalisations this year are very unlikely to surpass those from 2017/2018.
Image
Image
Quote Tweet
stuart mcdonald
@ActuaryByDay
Image
Flu arrived early and hit pretty hard this year. But it’s all over bar the shouting. Hospital admissions with flu have plummeted in the last 3 weeks. assets.publishing.service.gov.uk/government/upl
9
82