benjamin weber

building stuff, on leave from @princeton ECE

[benjamin weber @benguyweber]

@imrohan335 Me and you but I'm the corgi

[benjamin weber @benguyweber]

@imrohan335 @4ndyXu

[benjamin weber @benguyweber]

apple dev support:

Call us and talk to someone nice who can help you fix your issue within 30 minutes☺️

microsoft dev support:

𓋴𓅱𓃭𓆯𓅂 𓅓𓇌 𓂋𓇋𓂧𓂧𓃭𓅂 𓏏𓅱 𓂋𓅂𓎢𓅂𓇋𓆯𓅂 𓋴𓅲𓊪𓊪𓅱𓂋𓏏 𓅱𓈖 Verified Credentials 𓇋𓈖 𓃉𓇿𓃋 𓉔𓄿𓂋𓆯𓅂𓋴𓏏 𓅓𓅱𓅱𓈖𓋴

[benjamin weber @benguyweber]

i tried uninstalling edge from a thinkpad and it instantly bluescreened

[Satya Nadella @satyanadella]

Any pursuit of superintelligence has to be grounded in the core principle that if the AI we build is not helping humanity and under human control, it's not worth pursuing.

We also need to accelerate and spread the benefits of AI, such that they are diffused broadly across countries, communities, and companies. This requires a frontier ecosystem in which both closed and open-source models can thrive.

And for firms, it’s imperative that they retain full control over their unique and tacit knowledge. Every organization should be able to build its own continuous learning loop/hill climbing machine, without becoming dependent on any one model provider, and have the ability to embed its own knowledge into models and weights they control.

So, in this context, we welcome the research, focus, and deliberate pacing needed to get alignment right as the design goal. We also welcome ideas like "embedded evaluators" and the broader efforts to develop the mechanisms to make this more than just talk.

The key is that this cannot be controlled by a handful of entities, but must have broad representation across the ecosystem, countries, and fields, including academia.

This is the approach we are taking: broad access and choice at every layer of the AI stack; enterprise control of learning loops and models; and the “Code of Conduct” that underlies our own first party MAI models that we’ll publish tomorrow for public consultation.

[benjamin weber @benguyweber]

@baoskee bao bro I'm crine

[benjamin weber @benguyweber]

@bert_programmer i forgot about him I haven't used claude in terminal in a long time

[benjamin weber @benguyweber]

what is everyone's favorite little guy from the different coding agents? big fan of grok bot little guy customization

[benjamin weber @benguyweber]

there really should be gag orders on anthropic employees... like what the hell are we posting

[Evan Hubinger @EvanHub]

Jacob is correct here—we really do earnestly believe AI could kill all humans! I personally think it is >10% within the next decade. I believe Anthropic is trying its best, but we do not yet have a plan to solve alignment for superintelligence and are not clearly on track to.

[benjamin weber @benguyweber]

I was using Poke for a bit, it was working and then post acquisition it got worse

Tried Instinct and asked for it to look on amazon for some stuff I needed. 2/4 things were dead links or unavailable

No chance that people use this to book flights right?????

[Lei (:3 っ)∋ @Shlissika]

“5 likes isn’t a lot” ok imagine 5 seals

[benjamin weber @benguyweber]

@shreypandya @adisingh lga supremacy

[benjamin weber @benguyweber]

@NYCMayor can we please get a bus from queens to brooklyn. 1 hr commute for a 20/30 minute drive is not awesome :(

possible route attached

[benjamin weber @benguyweber]

@eosborn88 @saylor @saturn_credit @AdamBLiv @Strategy @strata_markets my goat

[benjamin weber @benguyweber]

@jasonzgao

[benjamin weber @benguyweber]

@imrohan335 save some for the rest of us

[benjamin weber @benguyweber]

@4ndyXu andy you're famous

[Andy Xu @4ndyXu]

3.75 years after chatgpt launched, I still find models to be bad collaborators. part of talking through an idea with smart human friends/mentors is that they are good at knowing when to *stop* talking. generally, when I ask a question, I want our conversation to be a DFS where you give me a high level A B C, then I can ask more clarifying questions to break down specific parts (maybe I will ask more questions about B, but then 5 minutes later we will go rabbithole on A).

otoh, when I ask chat/claude (esp in codex/cc), models always prefer to over rather than under explain. this means that density of information divided by time is quite low (often models go on tangents that idc about and generate neatly formatted multi-section essays for questions I want a few sentence answer to).

when humans write, we braindump all our ideas first, then cut and rephrase our writing repeatedly over time to make ideas succinct & elegant. I want models who take their "braindumps" and know how to condense them to the key things I care about (given all context they have about me) before responding.

[benjamin weber @benguyweber]

people have been kind of talking a lot about something that feels adjacent to this which is that models have increasingly started using more "jargon"?? I haven't looked much in to why this is happening to deeply but it drives me insane.

also you should try chatgpt voice if you haven't recently. I actually like it a lot, I tried it for the first time in probably over a year yesterday and it seemed to do better at the exact problem you are talking about here (the DFS and A B C).

[International Cyber Digest @IntCyberDigest]

‼️ BREAKING: An active npm supply chain attack has compromised at least 868 packages carrying over 2 billion monthly installs with a credential-stealing worm. Shai-Hulud is back.

It started with the compromise of the GitHub account of the maintainer behind keyv, a library with roughly 127 million weekly npm downloads.

A preinstall hook fires on npm install and drops a stealer that sweeps npm, GitHub, AWS, Kubernetes and Vault secrets, and then spreads to more maintainers.