kibae

> Opus 5.5 is the first Opus model to launch with a similar class of safeguards to Fable 5.1 on cybersecurity, biology, and distillation, all of which fall back to another model transparently.

This is where Chinese models are going to eat Anthropic's lunch.

Speculating here, but I assume researchers can make a reasonable estimate of the size of closed models based on factors like training time, training speed, and the number of tokens processed.

Also, Anthropic and OpenAI probably want to keep each other on their toes so they don’t end up on the wrong side of another Opus 4.6 / GPT-5.3-Codex situation, where one lab releases a model only for the other to drop a better one hours later.

Funnily enough that is against the Instagram ToS (https://help.instagram.com/termsofuse):

> 4.2 How You Can't Use Instagram. [...]

> You can’t sell, license, or purchase any account or data obtained from us or our Service, regardless of whether such data was obtained while logged-in to an Instagram account.

> This includes attempts to buy, sell, or transfer any aspect of your account (including your username); solicit, collect, or use login credentials or badges of other users; or request or collect Instagram usernames, passwords, or misappropriate access tokens.

There was a HN post that tested this hypothesis a few weeks ago: https://news.ycombinator.com/item?id=49010129

The training data would need to have a permissive license for this to be possible.

Cloudflare, Azure, AWS, and Google Cloud all have a similar uptick in reported errors around 7:30. I suspect an outage on Cloudflare or another load-bearing service cascaded through all the major cloud providers.

https://downdetector.com/status/cloudflare/

https://downdetector.com/status/windows-azure/

https://downdetector.com/status/aws-amazon-web-services/

https://downdetector.com/status/google-cloud/

Since most languages read from left to right, rightward movement tends to read as forward progression. So when showing a bicycle in side profile, having it face right feels more naturally like it’s moving forward.

The language selection is in the footer. Interestingly, the URL for English is the original post without en-US.

Using OS as an abbreviation for open-source and operating system in the same thread is hurting my brain.

Those numbers are incorrect unless they got a deal that's 100x cheaper than API pricing which is unlikely. They updated the original post with the correct costs.

The cost seems to be using the wrong symbol: ¢ vs $

Also the ability to uninstall bloatware like Chess. When I switched my browser away from Chrome I kept typing "c, h, return" in spotlight out of habit and kept opening Chess.

There seems to be an uptick around 1am on Polymarket.

https://polymarket.com/event/us-strikes-iran-by

There are mirrors on its' wikipedia page: https://en.wikipedia.org/wiki/Library_Genesis

When I was in high school, the physics teachers set up a telescope aimed at Venus as it crossed the Sun's radius. It wasn’t visible to the naked eye, but through the telescope, you could see Venus as a tiny black dot drifting across the field of view. It was fascinating to watch a planet move in a perfectly straight line along its orbit.

I switched to Brave last week after the whole Firefox fiasco. I installed uBlock Origin after there were some ads that got through.

e.g. on DuckDuckGo.

I always suspected that bots on Reddit were used to gain karma and then eventually sell the account, but maybe they're also being used for some kind of RLHF.

Pigeons were used in WW1 and WW2 for communication. Paddy the pigeon [0] flew 230 miles across the English channel to relay the success of the D-Day invasion.

[0] https://en.wikipedia.org/wiki/Paddy_(pigeon)

These days, I ask Claude/ChatGPT to create the regex and usually I know enough to be able to verify it. To double check, I'll start a new conversation and ask it what the regex does and verify it that way.

You can also ask it to create unit tests with edge cases. It might not catch every edge case, but usually it will create edge cases that you might not think of when writing unit tests yourself.

A note-taking app that is similar to Obsidian where I can link notes to each other.

Some differences include no naming rules (a lot of my note titles have a colon) except uniqueness, custom clusters so my knowledge graph is manageable.

I’m going to add different note types (i.e. not every note is .md, but have .csv and .ical as well) that can be expanded and linked within a note.

It’s written in React, hosted on AWS (so not an Electron app… yet), and CodeMirror for the editor.

I’ve spent probably 100 hours making it so far.

There's a clip of it on a MKBHD video on a different Disney product [here](https://youtu.be/1KEtxTQUzxY?t=282).

Tiny Habits by BJ Fogg might be a better book for you. The key to building habits is to start with a habit that is ridiculously small and once you feel like you've got that locked in, you can add another ridiculously small habit on top of it.

e.g. To build a flossing habit, floss just one tooth a day. Once you've got into the habit of pulling out floss and flossing just one tooth, you can move onto flossing two teeth and so on.

If at any point you feel resistance to doing the bigger habit you're trying to build up, you can always revert back to doing the initial tiny habit of flossing just one tooth.

Atomic Habits.

It's a simple book that can be summarized in an essay, but listening to it repeatedly while driving helps me keep my life on track.