chubby 🐼 🪭

[BridgeMind @bridgemindai]

I genuinely thought Claude Max 20x meant 20x the usage of Pro.

I have paid $200 a month for over a year. I had four Max subscriptions at one point. I run Claude Code with Fable 5 all day.

And I did not know until this week that the 20x only applies to the 5 hour window. The weekly limit has no multiplier at all.

I was misled by Anthropic.

[EnergyUp @EnergyUp_]

Neuroscientist Dr. Giedd tracked adolescent brain development for over 20 years.

Between ages 9 and 15, boys experience the second-largest brain growth spurt of their entire lives.
Gray matter thickens.
Neural pathways get rewired.

The emotional center grows faster than the control center.
He's not choosing chaos.
His brain is chaos.

[Elon Musk @elonmusk]

Extremely severe extinction events happen every 100M years or so and just switching to sustainable energy will not be enough to stop them.

Space satellites to control temperature and massive geo-engineering will be needed or it will be game over.

https://t.co/y4QwutjXce

[wd 🇵🇹🔺 @populartourist]

Is anyone using Hermes for complex code work using its native tools?

Honest question, because I let GPT 5.6 Sol Max work on a branch for 8 hours and left me with 900 warnings on a Rust repo.

Trying to understand the limitations here - not pulling the leg the amazing work Nous Research has done so far as it's my daily assistant.

[Qwen @Alibaba_Qwen]

⚡Meet Qwen3.8-Flash, a multimodal MoE and an early preview of the Qwen4 architecture, now open-weight!

The production version Qwen3.8-Flash will be available soon via QwenCloud API at just $ 0.16/1M input tokens and $ 0.47/1M output tokens.

125B parameters + 51B N-gram embeddings, with just 6B activated per token. Unmatched cost-efficiency.

What's new: 🥳
- Next architecture: GDN + QSA hybrid attention, Gated Residual, N-gram Embedding & Muon optimizer, serving as a precursor to the architecture used in Qwen4.
- Dramatically lower training and inference costs: trained at just 1/9 the cost of Qwen3.7-Plus, while outperforming it across the board with especially strong gains in coding and office tasks.
- Strong performance: scoring 58.7 on DeepSWE 1.1, 62.5 on SWE-bench Pro, 73.9 on CoWorkBench, 84.5 on AndroidWorld, and 95.7 on MathVision (with CI).
- 262K native context, extensible to 1M with YaRN.

We’re also releasing the weights for Qwen3.8-Flash-Next, giving the community an early look at the new architecture we’re exploring for Qwen4.🚀

We can't wait to see what you build with Qwen3.8-Flash!👀👇
- Blog: https://t.co/M5hYypFLgJ
- Technical Report: https://t.co/IF0gObIkQO
- Hugging Face: https://t.co/6ow8QVAABt
- ModelScope: https://t.co/tDOn2jNuFG

[chubby 🐼 🪭 @chubbypandafan]

@MiaAI_lab I thought you ment quality over a larger context window. 1m tks would be nice!

[chubby 🐼 🪭 @chubbypandafan]

This!

[Rob Moore @robprogressive]

Going to University in the UK could be a financial trap

We tell 18-year-olds to sign up for £50,000+ of debt for a degree in ‘Business Management’ that leads to a £24k admin job

Meanwhile, the 18-year-old who started a plumbing or electrics apprenticeship is making £50k+ by the time they’re 23

With zero debt & a skill that Claude can’t replace

The university obsession was a trap pushed to keep unemployment numbers looking low

[Alastair Hilton @London_W4]

This month’s Prime Minister Andy Burnham @andyburnham has just announced he’s increasing foreign aid to 0.7% of GDP, which is £21 billion per year. Let me put that in context for you. The average amount of income tax paid per year by basic rate taxpayers in the UK is £2,700. So that means that nearly 8 million taxpayers in the UK are working all year ONLY to give their wages to foreign governments. There are 30 million basic rate taxpayers in the UK, so one in four people that you see working in restaurants or pubs, nurses, cleaners and many more are paying all of their tax to foreign countries. Countries like Pakistan that can afford nuclear weapons, to India that can afford a space programme, to China that can afford to buy the world, to Somalia that doesn’t have a government, only warlords. Etc etc etc etc etc. One in eight British basic rate taxpayers are paying all of their money not to schools in this country. Not for hospitals in this country, nor to fix the roads, nor anything whatsoever in Britain. Every penny goes abroad. Eight million people here giving their hard earned money to the British government who then give every penny away. Every single penny. So, if you’ve ever wondered why things aren’t improving in this country and wonder what happens to the money you’ve made from your toil and your sleepless nights of wondering how you’ll pay the next bill, that’s what happens to it. 8 million of you are working for foreign governments. Think about that.

[Creative Deduction @CreativeDeduct]

“Give a man a fish, and you feed him for a day.

Teach a man to fish, and you feed him for a lifetime.

Steal a fish from one guy and give it to another - and keep doing that on a daily basis - and you'll make the first guy pissed off, but you'll make the second guy lazy and dependent on you. Then you can tell the second guy that the first guy is greedy for wanting to keep the fish he caught. Then the second guy will cheer for you to steal more fish. Then you can prohibit anyone from fishing without getting permission from you. Then you can expand the racket, stealing fish from more people and buying the loyalty of others. Then you can get the recipients of the stolen fish to act as your hired thugs. Then you can ... well, you know the rest.”

- Larken Rose

[Callum Lyon @CallumLyon]

Even an average salary in the UK isn't enough to survive anymore.

The average salary in the UK is apparently £39,039, that's around £2,427 take home a month after tax, national insurance, pension and student loan contributions.

The average rent in the UK is £1388 a month.

That's £1039 left just paying rent.

The average council tax is £200 a month.

That's £839 left.

The average gas and electric bill in the UK is £139 a month.

That's £700 left.

Then add £300 a month for car insurance and fuel.

That's £400 left.

Then add in £300 a month for food.

That's £100 left.

Phone bill £50

£50 left.

And theres still things to pay.

How did it get this bad?

[chubby 🐼 🪭 @chubbypandafan]

@MarioNawfal Bot should pick up his own box

[Russell Brand @rustyrockets]

Bloody hell. I’ve been grossly misinformed about Bolshevism.

[chubby 🐼 🪭 @chubbypandafan]

Ran your exact recipe today on a 4090 (same model, official Q4_K_M drafter, PR #27342 + a guard for the 24GB VRAM edge) and can confirm your numbers almost exactly — 73.6 t/s decode at n-max 3 vs your 75.7.

Two findings from a angle nobody's benching: we ran it inside a real agentic tool-loop (financial doc reconciliation, penny-exact answers) instead of raw prompts.

1. Quality: 2 full runs, every answer identical to our non-speculative gated baseline. Spec decode held up under multi-turn tool calling.
2. The catch: at thinking-mode sampling (temp 1.0 / top_p 0.95) drafter acceptance tanks — composition decode drops to 33–60 t/s, sometimes below no-spec, while doc-quoting bursts still hit 80–100. Your new --spec-draft-p-min is probably the fix for exactly this.

Also worth noting from your own table: MTP prefills ~40% faster at every depth — for prefill-heavy agent loops that claws back a lot of DFlash2's decode edge.

[chubby 🐼 🪭 @chubbypandafan]

@BGatesIsaPyscho

[chubby 🐼 🪭 @chubbypandafan]

@elshayib_ its not the same anymore

[Elshayib @elshayib_]

Anthropic is Fucked

Crazy amount of users canceling subscriptions right now.

[Tak 🦞 @cherry_mx_reds]

I have cancelled my Anthropic sub until things improve. I do not need my LLM telling me to go to bed, refusing to work because of the remaining context window and other bizarre behavior.

[chubby 🐼 🪭 @chubbypandafan]

@jun_song you have rich friends!

[chubby 🐼 🪭 @chubbypandafan]

@LottoLabs If money wasn't an issue then id pick the GPU

[chubby 🐼 🪭 @chubbypandafan]

@Paulrestore Those that put him there deserve to take the stand themselves. Good job for staying firm teach!