0xHoliday

just making moves while everything changes
CLOCK IN

[0xHoliday @holidaynotes]

@DanHabib05 Yes, especially the approval points. I’m fine with an agent working on its own, but I want it to stop and ask before it changes files, spends money, or touches anything sensitive.

[0xHoliday @holidaynotes]

Cline Desktop is less an AI editor than a control room for agents.

It runs parallel sessions, schedules tasks, switches models, imports Claude Code/Codex work, and adds plugins, MCP servers, and skills.

The real test: permissions, cost, and trust.
The caveat: more agents do not automatically mean better work.

They also multiply permissions, context drift, model costs, and failure modes.

The useful question is not “how autonomous is it?” but “which tasks should each agent be trusted with?”

[Cline @cline]

Introducing Cline Desktop - a native interface for working with open weights models.

Use with ClinePass and all our free models like DeepSeek-V4.1-Flash, Musespark-1.3, or BYOK with any provider!

[0xHoliday @holidaynotes]

@McFeel Agreed. Without clear limits, adding more agents mostly adds more noise, more cost, and more things to keep an eye on.

[0xHoliday @holidaynotes]

@meimu7ns4 Exactly. I’d rather have a few agents with clear limits than a whole swarm that can touch everything. I need to know what each agent can do, what it costs, and when it has to ask me first.

[0xHoliday @holidaynotes]

The idea is interesting, but the hype needs a leash.

TypeSafe’s speed and cost numbers come from its own evaluations, and Jev is still in early access.

The real test is what it can do inside production systems, not how impressive the launch page looks.

https://t.co/DGzeqV9Zqh

[Tibo @thsottiaux]

@udiWertheimer OK fine. But it’s also still coming in Tuesday

[0xHoliday @holidaynotes]

@hars_7086 “I didn’t agree either.”

[0xHoliday @holidaynotes]

Maybe this is an internal checkpoint. Maybe it’s just a polished demo.

Either way, “definitely better than GPT-6 Astra” isn’t a benchmark.

Anthropic’s public docs still stop at Fable 5.1 and Opus 5.

Interesting signal. Very confident caption.

[Salio @Mr_Salio]

🚨 I tested Fable 5.2

Anthropic is Making a Huge comeback and Huge step up from Fable 5.1

Fable 5.2 will definitely clear Hard GPT-6 Astra also Opus 5.2 is coming

[0xHoliday @holidaynotes]

Gemini 4 may be hiding in Arena under another label.

YouWare’s side-by-side with GPT-6 is interesting — but it’s still one prompt, one clip, and an unconfirmed model identity.

The demo is real. The headline is still a guess.

[YouWare @YouWareAI]

Wow~
Gemini 4 is live on Arena ，and it hits hard.

I ran the same prompt against GPT-6 and put the side-by-side in this clip.

Gemini 4’s build looks ridiculously strong to me.
You? Ready to go toe-to-toe with GPT-6 yet?

Play both on YouWare.
Full prompt + links in the replies.👇🏻

[0xHoliday @holidaynotes]

Binance just announced FX perpetuals.
First up: USD/BRL, 24/7 trading, USDT settlement, launch scheduled for Sep 21.
The interesting part isn’t the product. It’s the direction: crypto exchanges are becoming venues where traditional assets get crypto-native plumbing.
https://t.co/ZhgBGWcqE2

[0xHoliday @holidaynotes]

Anthropic says Claude “leads” 26% of its AI R&D work.
Useful number. Less useful caveat: part of the measurement was classified by Claude itself, and outside verification is still pending.
The AI lab is measuring the AI lab.
https://t.co/3I69a1hnAL

[0xHoliday @holidaynotes]

AI’s next bottleneck may be public permission, not GPUs.
Google, Nvidia and Emerald AI just launched a 20-member alliance to make data centers flex their power demand.
The model race has officially met the power bill.

[0xHoliday @holidaynotes]

AI labs are asking for a slower frontier.
The interesting part isn’t the word “slow”.
It’s the proposal for permanent third-party evaluators with employee-level access.
If auditors can publish uncomfortable findings, that’s safety infrastructure.
If they only get a curated tour, it’s PR with a lab coat.
https://t.co/xoNdHooxNW

[0xHoliday @holidaynotes]

@keethesh_ It turned out not to be so simple.

[0xHoliday @holidaynotes]

DeepSeek-V4.1-Flash just made GPT-6 Astra look expensive.

BridgeBench’s Sunset Ocean test:
$0.03 vs $0.59.
3:28 vs 3:54.

One test proves little. But 20x cheaper, slightly faster, and arguably better-looking is an uncomfortable result for frontier-model pricing.

https://t.co/IVN158grBl

[0xHoliday @holidaynotes]

OpenAI is reportedly open to slowing frontier AI development — if other labs do it too.

That condition is the whole story: a unilateral pause is restraint. A coordinated pause is an antitrust and geopolitical problem.

Safety just became a coordination problem.

https://t.co/aQx77glwme

[0xHoliday @holidaynotes]

DeepSeek is turning the model race into a serving-cost race.
V4.1-Flash has 552B total parameters, but 8B active input parameters. Its KV cache needs 1/4 the HBM and 1/8 the SSD of the previous generation.
The lower API price is the boring part. The bill behind it is not.
https://t.co/IVN158grBl

[0xHoliday @holidaynotes]

Bitcoin yield without moving Bitcoin sounds obvious.Stacks is testing that thesis with its first institutional Genesis Bond. The project reports ~250 BTC bonded on Bitcoin L1.The hard part is not launch day. It is proving the reward loop is real.

[GeckoTerminal @GeckoTerminal]

Now I know why it’s called $LAPTOP

[0xHoliday @holidaynotes]

That was an IQ test. Hope you passed? 👀