It’s been five years.
Five years of building.
Five years of believing.
Five years of learning the hard way that being early can look a whole lot like being wrong.
The technology was beautiful.
Maybe too beautiful.
@UsmanReads I don't like when AI is apologizing like that. It's meaningless.
It would be different if it's learning or adding some rule to avoid the error in the future, but just saying sorry is only wasting tokens.
Is cost per token even a relevant metric?
Not really.
What actually matters is cost per task:
cost per task = tokens used × price per token
A model cheaper per token can cost more if it uses more tokens or needs retries.
Unfortunately this is not tracked by most benchmarks.
That was the problem I had as well. But I went for building my own hardware solution. Lights up blue if idle/done, yellow while working, and red when permissions are requested:
My current workflow has 6+ tabs of Claude Code running at all times.
M pain: I'd leave a tab unattended, come back 20 minutes later, and realize Claude had been waiting for permission the whole time.
I also would have to navigate from tab to tab checking all. Felt broken.
🧪 Experimental: Use OpenCode with Claude Code, Codex, and Amp
- Universal coding agent control
- HTTP API for sandboxed agents
- OpenCode TUI, web UI, SDK
Available in Sandbox Agent SDK 0.1.6
@JustinJackBear@YouTube Great. Since you mentioned hosting IC nodes on consumer hardware, you might want to look into Badlands, which was a proposal to do just that with a subnet or parallel, interoperable network.
Unfortunately it didn't happen but might be interesting to you. https://t.co/MWkwIw89Ge
@ErikSchluntz@tomilton I would expect this case to be covered by https://t.co/sR5lneQF5V
My main concert is that the idle notification only shows up "after 60+ seconds of idle time". But I guess I just have to run some tests and see how it feels.
Chinese scientists have developed,
The best shortest-path algorithm in 41 years!
A team from Tsinghua University has broken Dijkstra's "sorting barrier" - the first improvement since 1984.
Just use for a world-map 🤯
Paper - https://t.co/0AhR5O7vl4
https://t.co/a9KMVRuYGx
A new DFINITY white paper proposing how #Mission70 (to reduce ICP inflation 70% during 2026) can be won, shall be published next Wednesday, 14th Jan. Soon after we will create an NNS proposal. Forward ICP 🔥
I have been having this problem with AI and it's magnified on X. Everyone is constantly talking about how amazing their experience is, yet here I am with skill issues never seeing the dividends these people get. I find that I fight a ton with incorrect answers, not in syntax, but in direction. Again I can admit skill issues. At the same time, I don't really care if I type every letter. I love LSPs and autocompletes because I can output faster when I am in the zone. So if I can extend that experience, I'll be so excited.
So I decided maybe I need an AI experience that is more my speed. Something in which compliments my lack of skill issues in programming. Something that could make me 50% faster, maybe even 2x faster while not compromising actual quality of any kind.
This is my first iteration. This is 99.
This is very alpha level implementation, but it's ready for some basic test driving and feedback. If you have any of the same feelings as me, check out 99.
https://t.co/JapGB7vxKY
Hate is not a crime. Emotions, thoughts and words, however repugnant to some, must be free in a free society. Actions are what laws and enforcement need to constrain.