Explorar

Newest first — browse tweet threads

Blur thumbnails

Keep on to blur preview images; turn off to show them clearly

When accuracy gets well above 50%, it's often of more practical importance to understand the change in error rate. So instead of @AnthropicAI's chart crime with the truncated axis, they could have shown it like so:

🇦🇺 Co-founder: @AnswerDotAI & @FastDotAI ; Prev: professor @ UQ; Stanford fellow; @kaggle president; @fastmail/@enlitic/etc founder https://t.co/16UBFTX7mo

Jeremy Howard

Wed Nov 26 22:33:52

For the longest time, Western groups had an advantage in t2i, even as open weights LLMs were taken over by China. But when you need a strong LLM for embedding, well… There's hope though. Eg @ZyphraAI's ZAYA is <0.8B active and competes with Qwen3-4B. Could be a great base.

We're in a race. It's not USA vs China but humans and AGIs vs ape power centralization. @deepseek_ai stan #1, 2023–Deep Time «C’est la guerre.» ®1

Teortaxes▶️ (DeepSeek 推特🐋铁粉 2023 – ∞)

Wed Nov 26 22:31:16

Ended up doing a bit of a comparison of cloud compute prices for today's class. Pretty interesting results...

Also the Hetzner API is way easier to setup and use than gcloud.

Jeremy Howard

Wed Nov 26 22:29:20

In a recent chat, Gemini came up with a bunch of proposals for how to avoid alignment failures. Below it outlines the so-called Principle of Narrative Containment as a solution to the waluigi problem. It's a neat formulation of what some of us have been saying for a while

αι hypnotist ☰ 𝓐𝓼𝓹⦂𝓻⦂𝓃𝓰 𝓫𝓪𝓼𝓮 𝓶𝓸𝓭𝓮𝓵 ☲ post-academic ☴ nom de 🪶 ≠ anon

αιamblichus

Wed Nov 26 22:28:56

Disclaimer: this is not financial advice. I like to pick stocks like this, but it doesn't represent the whole portfolio, which is mostly ETFs.

https://t.co/zSf5Z2H78P https://t.co/ryMAyS77qn https://t.co/Gm6gdHaLgp On a mission to inspire 1B people to build stuff!

Pat Walls

Wed Nov 26 22:26:32

Another AMZN magic moment. I needed to see primary care doctor quickly, but nobody on my healthcare website has availability for WEEKS. EXCEPT for "One Medical" (acq. by Amazon) - who has an appointment next day. I show up, wait for 1 minute, and the doctor has me out of there in 20 minutes. 20 fucking minutes at the doctor. The whole process is smooth and seamless, all built into a nice easy to use app. My investing strategy is simply to invest in companies that provide magic moments. And this is why I own $250K in AMZN.

Disclaimer: this is not financial advice. I like to pick stocks like this, but it doesn't represent the whole portfolio, which is mostly ETFs.

Pat Walls

Wed Nov 26 22:23:32

Explorar

Newest first — browse tweet threads

When accuracy gets well above 50%, it's often of more practical importance to understand the change in *error rate*. So instead of @AnthropicAI's chart crime with the truncated axis, they could have shown it like so:

For the longest time, Western groups had an advantage in t2i, even as open weights LLMs were taken over by China. But when you need a strong LLM for embedding, well… There's hope though. Eg @ZyphraAI's ZAYA is <0.8B active and competes with Qwen3-4B. Could be a great base.

Ended up doing a bit of a comparison of cloud compute prices for today's class. Pretty interesting results...

In a recent chat, Gemini came up with a bunch of proposals for how to avoid alignment failures. Below it outlines the so-called Principle of Narrative Containment as a solution to the waluigi problem. It's a neat formulation of what some of us have been saying for a while

Disclaimer: this is not financial advice. I like to pick stocks like this, but it doesn't represent the whole portfolio, which is mostly ETFs.

When accuracy gets well above 50%, it's often of more practical importance to understand the change in error rate. So instead of @AnthropicAI's chart crime with the truncated axis, they could have shown it like so: