Show newer

Just tried to find the formula for a rotation matrix in n-dimensional space and ChatGPT, Bard, and Bing all gave me the wrong answer.

Joe boosted

Long-sought universal #flu #vaccine: #mRNA-based candidate enters #clinicaltrial
An mRNA-based flu vaccine designed to offer long-lasting protection against a broad range of #influenza viruses is now in a phase I clinical trial, #NIH announced. The phase I trial will test safety and efficacy in a small number of people. arstechnica.com/health/2023/05

The new paper demonstrates that LLMs are terrible at autonomous planning but are helpful in improving the results of heurisitic planners.

Show thread

Anyone remember Microsoft's "Sparks of AGI" paper that claimed that GPT-4 shows hints of human-like intelligence? Well, if you consider planning to be part of intelligence the authors might need to recalibrate their conclusions. arxiv.org/pdf/2305.15771.pdf

Joe boosted

Big claim that shingles vaccine prevents dementia. This will be huge if it holds up.

Sadly for us men the effect seems to be mainly seen in women

medrxiv.org/content/10.1101/20

Joe boosted

Come the fall, seniors in the US will be urged to get vaccinated against #RSV, a respiratory pathogen that causes about as much illness in this group as does flu. But RSV experts worry neither seniors nor their doctors understand the risks RSV poses. statnews.com/2023/05/26/rsv-va

Joe boosted

The loss of access to Twitter data under the new pricing plan was a dire blow to our research at the UW-CIP.

Now being forced to delete the data that dozens of people spent years collecting, curating, and studying—outrageous. It’s like burning lab notebooks, an attack on science for the sake of pure malice.

inews.co.uk/news/twitter-resea

Quite a few of the commenters here are equating Bender's thought experiment with Searle's famous Chinese Room experiment. I think this is incorrect .

I think Bender is trying to answer the question: is it possible to derive the meaning of symbols without understanding the external representation of those symbols? Suppose you locked a newborn baby in a Skinner box (not recommended) and only allowed the child to listen to English audiobooks. Would that child learn to understand the meaning of the words it hears?

Searle's thought experiment, on the other hand, is designed to answer the question of whether is it possible to assign language understanding to a program that can accurately translate a language.

Joe boosted

One of the decisive moments in my understanding of #LLMs and their limitations was when, last autumn, @emilymbender walked me through her Thai Library thought experiment.

She's now written it up as a Medium post, and you can read it here. The value comes from really pondering the question she poses, so take the time to think about it. What would YOU do in the situation she outlines?

medium.com/@emilymenonbender/t

Joe boosted

We can expect to see a lot of this. Prof turns down a tenured job at University of Texas - to join their conservative institute no less - because of the legislature politicizing higher ed and uncertainty about tenure. Regardless of outcome, the government’s hostility to its state universities will leave a mark. texastribune.org/2023/05/23/ci

Joe boosted

Lessons, living with parents with dementia: when an app on their phone spontaneously changes it's icon, it's a fucking disaster.

Some Google engineer launched a feature that got them a promotion and now my mom can't use her phone anymore. Silicon Valley is incapable of imagining a user who isn't a 27 year old white man.

"Look at our one pager on plans for dealing with imaginary unrealized dangerous superintelligent AI while ignoring any of the problems with our current products!" - OpenAI. openai.com/blog/governance-of-

Joe boosted

AI generated fake news briefly caused markets to fall, but the only thing we have to worry about regulating is the singularity, right?

insider.com/ai-generated-hoax-

Joe boosted

Digital information technologies are having unprecedented effects on society. Both their implementations and consequences are extremely difficult to monitor, and efforts to do so are often impeded by the technology companies that deploy them.

Today we have piece out in which we make the case for an intergovernmental body, analogous to the IPCC, with the role of collating information about implementations and consequences, benefits and harms of these systems.

nature.com/articles/d41586-023

Entertaining (at least for me) physics discussion of why the problem with SUV/pedestrian accidents isn't due to the mass difference relative to sedans but the height of the SUV. SUV/sedan collisions are a different matter.

docs.google.com/document/d/1nE

Joe boosted

Looks like “a [completely uninterpretable] deep neural network [with substantial unreported hyperparameter and architecture tuning] reproduced [some aspects of] our brain data”

has replaced

“a simple computational model capturing our proposed mechanism reproduced our brain data”

as the new figure 7 strategy for high-profile neuroscience papers

Show older
Qoto Mastodon

QOTO: Question Others to Teach Ourselves
An inclusive, Academic Freedom, instance
All cultures welcome.
Hate speech and harassment strictly forbidden.