At 17,000 tokens per second, what won’t be generated in real-time?
02.03.2026 19:01
👍 0
🔁 0
💬 0
📌 0
At 17,000 tokens per second, what won’t be generated in real-time?
We’re building an agent inside of Emra, so you’re able to ask for meeting summaries, get context about your conversations, and organize your scattered thoughts.
Legacy hardware will be no match for AI.
It’ll feel like Iron Man’s phone taking over the courtroom TV screen.
Deleted X account. Will post on website and youtube, and then bluesky and threads moving forward.
iPhone on aggressive "No Notifications Mode" and "Auto Decline Unknown Callers"
MacOS 26 Tahoe still shows a call screening box every time a telemarketer calls. Super invasive and flow breaking.
For clarification, they were using Cluely to talk to me.
I'm hiring a founding engineer, and a candidate was using AI during the intro interview.
When I realized my heart sank.
It was dishonest, but it was more sad that they felt desperate enough that they couldn't do it on their own (despite their impressive CV).
The current Google Fonts preview text is from the Universal Declaration of Human Rights (UDHR)
Turns out the Apple design team copied my jailbroken iPhone from 2010
Recently refreshed my website, check it out for some fun reads and interactivity.
Apple, it's not a want. It's a need.
Okay apple, here's what you need to do next.
Hint: this is not what these responses should actually look like.
You know what? Google actually cooked with this new homepage.
thrds likes this one.
You know, it's actually crazy that Apple neglected the primary thing that Siri needs.
Chat history, search history, some sort of home for Siri.
I certainly hope it’s clear that if you’re building a product to reduce friction while using cursor + a LLM, your product will probably only have a market for about 12 months max.
Get used to doing hard things.
People keep saying SaaS is over.
But I think the role of who is building these tools and services is going to drastically change. It will start to exist more at smaller agency and freelancer levels in ways that it wasn't able to before.
My whole life I've been told how hard it is to do the things I want to do.
And I can confirm, things are hard, but why in the world would I want to live my life on the wrong side of it?
If you get good at hard things, hard things become easy.
wild to me that TT and YT can do it better
X, BS, and Thrds all have completely destroyed the quality of feeds compared to what we had on Tweety not even 3 years ago. The plot is lost.
The algorithms pump brainrot and attention baiting.
I'm sorry discover feeds, I am now only using following or curated feeds.
I see more politics here than on X.
As far as I can tell, no amount of tuning of the BlueSky discover feed can actually remove politics.
I aggressively say "See less of this" and have the topic off in settings. I've blocked keywords, and even use custom feeds (which aren't great for discovery).
Okay pretty sure I need all AI models to support 1M tokens from here on out.
You know how airpods automatically hijack your microphone input on Mac?
I have a repeating shell command running inside Better Touch Tool that checks every few seconds to see if my airpods are connected. If so it forces microphone input from my AirPods to my lapel microphone.
Yeah, and I actively suggest "more of this" "less of this" but it doesn't seem to be too effective.
I would much rather spend my time here than X, but I don’t see relevant content here - not sure how to better tune my feed - do I need to make the better performing feed algorithm? Is that it?
Discoverability needs to get better in bluesky
I'd bet all the people building Loom clones are only doing it now because @thewebprodigies dropped a 16 hour tutorial in October.
I've seen the team at OpenAI talk about how they know they're bad at naming conventions, so I'd assumed it was a problem they were working on solving...
but I can't believe that they truly just went from 4.5 to 4.1.
Guys, it's not hard. Just increment the number.
