Philip Bell

Progressive technology and education | previously @georgiatech @harvard

[Philip Bell @PhilipfvBell]

Given the recent ban of Claude Fable 5 it is worth thinking about parallels between AI and GPS.

When the US invented GPS they allowed other countries to use it, but only with ‘Selective Availability’ (SA). This intentionally degraded the accuracy from about 20 metres to 100 metres, making it much less useful in military operations. In essence, the US was exporting a technology which was deliberately made worse for foreign nations.

President Clinton stopped Selective availability in May 2000 because they developed the ability to selectively deny GPS signals on a regional basis, as well as the launch of the European competitor system (Galileo).

Are there lessons to learn for AI sovereignty?

[Philip Bell @PhilipfvBell]

@MokshaMani Interesting that’s a good shout Mani

[Philip Bell @PhilipfvBell]

I find AI helpful, but I deliberately have slots in the day where I read and write without it. This is partly because I think cognitive independence matters personally and politically.

Does anyone have good ways of using AI while staying independent?

[Andrew O'Brien @ADOBrien88]

Best discussion of AI and its (possible) potential (or non-potential) I have read so far 👇

[David Edgerton @DEHEdgerton]

Had a very interesting conversation with @PhilipfvBell on how to think historically about AI and 'technology' more generally. @KingsCHoSTM @kingshistory https://t.co/77CvpqppT9

[Philip Bell @PhilipfvBell]

I spoke to David Edgerton about how analogies are misused in understanding AI and why he thinks we are not able to have 'sensible, grown-up' conversations about our material reality if we ignore actual machines, processes and materials.

Check it out:
https://t.co/lD3rVvmZyW

[Philip Bell @PhilipfvBell]

Simone Weil said attention was the highest form of generosity. AI can give unlimited attention 24/7 so is it limitlessly generous? Or does Attention require something AI doesn’t have?

[Philip Bell @PhilipfvBell]

Should governments buy AI stocks on behalf of their citizens?

As a hedge against all economic value accruing to a small number of investors, Casey, Roy and Rockall argue states should take a stake across AI companies (an ‘AI Bond’). Then populations share the upside if AI goes to the moon.

Is this AI bond idea sensible?

[Philip Bell @PhilipfvBell]

‘Positive Alignment’ means building AI optimised to support short term and long term flourishing. This is quite different from social media or standard consumer software, which are optimised for clicks.

But the reasons existing software is optimised for clicks haven’t disappeared. Therefore, what is the theory of change that can make positive alignment happen?

https://t.co/X3bgedMGv3

[Philip Bell @PhilipfvBell]

@lulumeservey Not sure I get your point! People have always had a lot of gear

[Philip Bell @PhilipfvBell]

Thought experiment.

Hamlet said: "thus conscience does make cowards of us all."

If a superintelligence had a deep and comprehensive understanding of the effects of its actions, would this lead to paralysis? Humans act partly because we are irrationally optimistic. If a superintelligent AI did not have this irrational ignorance and uses some form of constitution for alignment, might it tend towards inaction?

[Philip Bell @PhilipfvBell]

Why is it that LLMs can't count? LLMs are very powerful but fail in unexpected ways. According to researchers at Google there are three main reasons that they struggle to count.

The representation of the text they create is 'blurry'
Numbers at the end of a sequence are drowned out ('Over Squashing')
Probability Calculation simplifies numbers

The Google researchers note that there is a lot we still need to learn from this jagged intelligence.

Full article here: https://t.co/e3rTAQdOQX

[Philip Bell @PhilipfvBell]

@a_bergeaud @erikbryn DeepMind is in the UK though right?

[Philip Bell @PhilipfvBell]

Could AI give us more choice about the speed at which we live our lives?

It was a real pleasure to talk to @nicklaslundblad about this and how it might affect human identity.

https://t.co/VgeO8G3iJN

[Philip Bell @PhilipfvBell]

@adam_tooze this might interest you. I have taken your LRB article argument and extended it by looking at the technicalities of the GPU and the transformer architecture using @sarahookr's wonderful article 'The Hardware Lottery'

[Philip Bell @PhilipfvBell]

The human brain is roughly a million times more energy-efficient per bit of sensory information processed than current frontier AI models. A child learns language from roughly 10⁷–10⁸ words of input by age 10. GPT-4 class models were trained on ~10¹³ tokens. This is a 100,000× gap in data efficiency.

How did we end up with AI that is incredibly powerful but inefficient?

Fracked gas, quantitative easing and the GPU all played a role.

https://t.co/xWRmPDbV3O

[Philip Bell @PhilipfvBell]

It was a real pleasure to speak to @carlbfrey at the @oiioxford, University of Oxford and author of How Progress Ends.

He makes a really powerful case that technology doesn't guarantee progress. The same society can thrive in one technological era and stagnate in the next if its institutions don't adapt. Europe caught up in mass production but has struggled in digital. Are our current institutions suited to emerging technologies?

https://t.co/xg3wYJDacS

[Philip Bell @PhilipfvBell]

@Richie_Allen Agreed

[Philip Bell @PhilipfvBell]

@birchlse Would you consider starting a humanities AI org?

[Philip Bell @PhilipfvBell]

Given Mark Carney’s recent speech about the rupture in US-led globalization, this conversation with Hamish feels more relevant than ever. He has a really interesting perspective on how middle powers should be pursuing digital sovereignty through middleware. They should primarily pursue interdependence not independence.

https://t.co/7YJ6dQnDwI