Marina

[Marina @just4masha]

Толерантность достигнет такого уровня, что умным людям запретят думать, чтобы не оскорблять идиотов.

[Restoring Your Faith in Humanity @HumanityChad]

Human chain saved a dog and became a statue ❤️

[𝕐o̴g̴ @Yoda4ever]

If it didn't move, I would have thought it was just a plushie..🐹🧸😅

[Adam Grant @AdamMGrant]

Dealing with difficult people may be hazardous to your health.

Data: Each additional "hassler" in your network predicts 1.5% faster biological aging—and worse health a year later. Non-spouse family hasslers are the most problematic.

Some ties aren't worth the toll they take.

[Luke Gromen @LukeGromen]

“It was just the prelude… Where they burn books, they will ultimately burn people too.”

Heinrich Heine (1797-1856)

[Yet another commodity guy @tleilax___]

The technical report is worth a read if you are interested in LLMs.

[Ann Nguyen @ann_nnng]

try it https://t.co/0tnI1OV2w0

[Kevin W. @Brink_Thinker]

How would you have handled it?🤔

[Marina @just4masha]

@kadmitriev 🤣

[Minh Do @minhsmind]

Such a cool ad! Brilliant concept and execution.

[Marina @just4masha]

@uk_wolfspitz @PeruginiNic +1

[Marina @just4masha]

@DD_Geopolitics And she still cannot speak proper english

[Arnaud Bertrand @RnaudBertrand]

Trump right now in his live address: "We're going to hit [Iran] extremely hard over the next 2 to 3 weeks, we're going to bring them back to the stone ages where they belong."

Pure savagery. And textbook genocidal: saying the Iranian people "belong" in the stone ages means he's targeting them as a people, which is the definition of genocidal intent.

That's where letting Gaza happen without consequences gets you...

Also pretty ironical to call others primitive while sounding like a barbarian king on bath salts.

[FT Energy @ftenergy]

Traders placed $580mn oil bet ahead of Trump post on Iran talks https://t.co/ekiQDApJhV

[Marina @just4masha]

@kadmitriev 🤣🤣

[Marina @just4masha]

@AgelNbres @IliaBouchouev Is it a what’s app group?

[zerohedge @zerohedge]

1BN barrels in OECD commercial crude stocks
1.2BN barrels in IEA strategic oil reserves
1.5BN barrels in Chinese oil reserves
400MM barrels in on water storage

0 LNG in strategic reserve storage.

The LNG situation is far more dire than oil

[Fatih Birol @fbirol]

Following @IEA Member countries' decision to release 400 mln barrels of oil stocks to counter disruptions, initial volumes have already been made available

Thank you to countries for their stock contributions & to Canada & Mexico for increased production: https://t.co/CjYXygzQpG

[Adam Tooze @adam_tooze]

"... in this case, actually closing a kill chain" ... CLOSING A KILL CHAIN!

[Chris Laub @ChrisLaubAI]

🚨 BREAKING: A Google researcher and a Turing Award winner just published a paper that exposes the real crisis in AI.

It's not training. It's inference. And the hardware we're using was never designed for it.

The paper is by Xiaoyu Ma and David Patterson. Accepted by IEEE Computer, 2026.

No hype. No product launch. Just a cold breakdown of why serving LLMs is fundamentally broken at the hardware level.

The core argument is brutal:
→ GPU FLOPS grew 80X from 2012 to 2022
→ Memory bandwidth grew only 17X in that same period
→ HBM costs per GB are going UP, not down
→ The Decode phase is memory-bound, not compute-bound
→ We're building inference on chips designed for training

Here's the wildest part:

OpenAI lost roughly $5B on $3.7B in revenue. The bottleneck isn't model quality. It's the cost of serving every single token to every single user. Inference is bleeding these companies dry.

And five trends are making it worse simultaneously:
→ MoE models like DeepSeek-V3 with 256 experts exploding memory
→ Reasoning models generating massive thought chains before answering
→ Multimodal inputs (image, audio, video) dwarfing text
→ Long-context windows straining KV caches
→ RAG pipelines injecting more context per request

Their four proposed hardware shifts:
→ High Bandwidth Flash: 512GB stacks at HBM-level bandwidth, 10X more memory per node
→ Processing-Near-Memory: logic dies placed next to memory, not on the same chip
→ 3D Memory-Logic Stacking: vertical connections delivering 2-3X lower power than HBM
→ Low-Latency Interconnect: fewer hops, in-network compute, SRAM packet buffers

Companies that tried SRAM-only chips like Cerebras and Groq already failed and had to add DRAM back.

This paper doesn't sell a product. It maps the entire hardware bottleneck and says: the industry is solving the wrong problem.

Paper dropped January 2026. Link in the first comment 👇

[MP Arizona☀️🏳️‍🌈💙🌵🐕🐕‍🦺🫂💦🏜🐟🌴🎙🌎🌻♍️🌊 @AzPetrich]

Look at the picture below. This is a three million dollar missile from a two hundred million dollar jet that costs twenty thousand an hour to fly. They use it to destroy people who live on ten dollars a day.

The world is truly sick.