Tl;DR: It's not a new organization, or non-profit institute. It's a Substack/blog.
>
DMI is a platform for researchers and thinkers from across Google DeepMind, Google, and the wider global research community to work on and publish creative, deeply informed ideas about a world with AGI. They will not always agree, and they will likely change their minds, as more data and information comes to light at the fast-moving frontier. We created this institute precisely because broad-based intellectual discussion and debate are required to arrive at a consensus about how to address the challenges and opportunities we face as a society.
(Hiding chains of thought is one of the most harmful things I think OAI/Anthr. has done, altough Google Gemini also encrypts its chains of thought, so it's interesting to see the DeepMind Institute crticising this.)
My read is that they're criticizing moving the reasoning (more) into the latent space, such that not even researchers/safety-evaluators can actually see what's going on.
Of course this still then has a 'trust me, bro' feel to it, but making it outright impossible to reasonably inspect the CoT is a significantly larger and riskier step, IMO.
ipsum2 · · focus · HN ↗
> DMI is a platform for researchers and thinkers from across Google DeepMind, Google, and the wider global research community to work on and publish creative, deeply informed ideas about a world with AGI. They will not always agree, and they will likely change their minds, as more data and information comes to light at the fast-moving frontier. We created this institute precisely because broad-based intellectual discussion and debate are required to arrive at a consensus about how to address the challenges and opportunities we face as a society.
trollbridge · · focus · HN ↗
<a href="https://institute.deepmind.com/essays/the-case-for-reasoning-transparency/" rel="nofollow">https://institute.deepmind.com/essays/the-case-for-reasoning...
(Hiding chains of thought is one of the most harmful things I think OAI/Anthr. has done, altough Google Gemini also encrypts its chains of thought, so it's interesting to see the DeepMind Institute crticising this.)
ppaattrriicckk · · focus · HN ↗
Of course this still then has a 'trust me, bro' feel to it, but making it outright impossible to reasonably inspect the CoT is a significantly larger and riskier step, IMO.
trollbridge · · focus · HN ↗
Encrypting CoT is just plain nuts yet we’ve all accepted it as normal.