Peter Eckersley

🔬 AI safety & transformations of capitalism @AIObjectives. Prev. led tech @EFF & research @PartnershipAI. Cofounder @letsencrypt #certbot & @privacybadger

[Peter Eckersley @pde33]

@NickPinkston A very interesting finding... though that visualisation is a bit unhelpful, it seems to have gray scales (maybe slightly different hues) in 4 different places?

[Peter Eckersley @pde33]

The probability distribution of all art nouveau images!

[Cosmic Spring JWST @CosmicSprngJWST]

You can see Earendel as it's lensed by the cluster, WHL0137-08, in the center of the image! Here is a zoom-in on the star itself! Stay tuned for an update from the science team!

[Peter Eckersley @pde33]

We are so lucky to have such a flukishly, magnificently lensed view of such a distant star as Earendel! Eagerly waiting for spectroscopy...

[Cosmic Spring JWST @CosmicSprngJWST]

There she is!!

[Peter Eckersley @pde33]

Perhaps these are hangovers from the Great Resignation and workforce changes during the pandemic. Or it's that paths of bureaucratic dependencies can become congested in the same way that supply chains did. Are there types of systemic failure we should be fearing right now?

[Peter Eckersley @pde33]

There some strange travel crises in the world these days. I've seen friends and colleagues affected by a couple of them, but are there others? 1/n

[Peter Eckersley @pde33]

There's an airline luggage handling crisis that sounds close to a positive feedback loop tipping point https://t.co/olM08upesv 3/n

[Peter Eckersley @pde33]

The abyssal mechanosphere... h/t @elizeyaquate

[Peter Eckersley @pde33]

@jasonbaumgartne If you really want something fancy on your desktop, wait for a 4090...

[Peter Eckersley @pde33]

@jasonbaumgartne If your question is "how can I train the most/largest models the fastest for under 10k", cloud services are the answer. An A100 costs ~10K, but you can rent them from (say) Google for under $3 / hour. Only worth buying if you have very consistent workload.

[Peter Eckersley @pde33]

@SophiaIgnatidou What's a piece of art that you find beautiful?

[Peter Eckersley @pde33]

So generative models dream up objects like this at the drop of a whimsical hat. Perhaps one day there will be a way to make them, so that our world is filled with such things?

[AI Objectives Institute @AIObjectives]

We need a logo for the AI objectives institute... a thread 🧵to pick one! Retweets and likes are votes :)

[Peter Eckersley @pde33]

@mer__edith @gregmepstein There are obviously huge societal, ethics and safety questions about Replika's product, but I don't think this thread has a full model of them. This study is possibly helpful too https://t.co/338oo9UQ7A

[Peter Eckersley @pde33]

@mer__edith @gregmepstein Perhaps take a look at forums where their users post. It seems there are also millions of women who are lonely and into forming relationships with chatbots...

[Peter Eckersley @pde33]

@gregmepstein @mer__edith What makes you think that Replika's userbase is all or mostly men?

[Peter Eckersley @pde33]

Speaking of which, there's also a great new paper from Stephanie Lin, Jacob Hilton and @OwainEvans_UK on calibration in words rather than logits, an approach that should really be in future versions of BigBench... https://t.co/ujtIO1LROD

[Peter Eckersley @pde33]

Excited to have worked on a little corner of this grand, ambitious project! @machinaut @realSharonZhou & I added some simple calibration measures to BigBench. ML models are usually overconfident, it's past time to start measuring & mitigating that!

[Jascha Sohl-Dickstein @jaschasd]

After 2 years of work by 442 contributors across 132 institutions, I am thrilled to announce that the https://t.co/wezEGzDEHt paper is now live: https://t.co/4Yg36EB9Ru. BIG-bench consists of 204 diverse tasks to measure and extrapolate the capabilities of large language models.

[Peter Eckersley @pde33]

I now think I was wrong: laypeople will be persuaded, but the poor software will be stuck arguing with ML and AI ethics researchers who believe it is all smoke, mirrors, zombies and lookup tables. 2/2

[Peter Eckersley @pde33]

I used to think that laypeople would firmly believe that software can't be conscious, until long after conscious AI was present to argue with them. 1/2