Meat Stack (3/3) retweeted
Replying to @tim_cook
@Alwaleed_Talal from @menlocollege was the first one who “believed” in Apple only if Steve Jobs was bought back, when the mothership was sinking to $7 in 1996…
he was the contrarian investor (probably influenced by Sandy Weill) who re-surfaced it !
@WarrenBuffett is a late coming investor who has since mostly withdrawn.
Apple continues as a cult with $70B iPhone cash cow sales with $70B Accounts Payable on a rolling horizon basis running quarter to quarter…
There’s also an injunction coming up from @Qualcomm over israeli Stroj reverse engineering the @Snapdragon
@Bajarin @YahooFinance
Meat Stack (3/3) retweeted
[기사] #방탄소년단, '마이크 드롭' 뮤직비디오 16억 뷰 돌파..팀 통산 4번째
m.entertain.naver.com/home/a…
Meat Stack (3/3) retweeted
He ain’t say which dad he was
'Rich Dad Poor Dad' self-help author Robert Kiyosaki is $1.2 billion in debt: report trib.al/b2DKFjZ
Meat Stack (3/3) retweeted
'Rich Dad Poor Dad' self-help author Robert Kiyosaki is $1.2 billion in debt: report trib.al/b2DKFjZ
kid benched casemiro vs brazil at a world cup why would he wait til 2027
🚨🔵⚪️ Ayyoub Bouaddi and his camp have now reached an agreement on personal terms with Manchester City.
#MCFC expected to seal the deal with Lille as next step, already advanced as Bouaddi preference would be to join City NOW — rather than in 2027.
25
Meat Stack (3/3) retweeted
🇺🇸 TAX: New York City's new pied-à-terre tax could apply to over 31,000 second homes, more than triple initial estimates.
Opened this dataset expecting a spreadsheet of standards and got a genuine computable graph instead!
1,590 concepts, 3,221 connections across 8 subjects. math, science, computing, life skills, anchored standard by standard in both the US and UK curricula.
Every concept is structured JSON with an age band and the actual evidence a child has to show to demonstrate mastery, not just a vague standard description.
Every prerequisite link is marked hard or soft with a written rationale attached explaining why the connection exists.
Assembled from 7 curriculum frameworks using AI agents, then reviewed by an actual team before release, which is the part that made me trust it more than I expected to.
It's a real DAG. You can tap any concept and see everything that has to be mastered before it.
Genuinely useful if you're building anything in edtech, adaptive learning, or assessment tooling.
Fully open source now. Free to build on.
Follow @neil_xbt for more.
Meat Stack (3/3) retweeted
T3MP3ST
A multi-agent offensive-security framework, built to turn the AI coding agent you already run into a zero-day hunter.
Credit/Source: github.com/elder-plinius/T3M…
Ok so GLM-5.2 gets hyped, China stock $ZHIPU gets listed on #Hyperliquid, and you guys not digging into the narrative?
A 744B-ish MoE model, only ~40B active per token, MIT open weights, 1M context, 131K max output, trained on Huawei Ascend chips with no Nvidia.
Launched right after Fable/Mythos got cut off globally from the US export-control mess.
Now it's the top open-weight model, and close enough to frontier closed models that the pricing gap looking kinda absurd.
– $1.40/M input, $4.40/M output, $0.26/M cached
– GPT-5.5 is around $5/$30 and Opus 4.8 is around $5/$25.
→ Near-frontier coding performance at ~1/5 to ~1/6 the cost. So the narrative could get pretty big.
Decentralized AI infra finally has an actual problem to solve:
– closed models have geopolitical kill switches
– open weights can't be turned off
– self-hosting is expensive
Some crypto projects are sliding into that stack:
– @ionet | $IO : day-0 GLM-5.2 launch partner. Long-context inference and self-hosting a model this size need distributed GPU supply.
– @PhalaNetwork | $Phala : running GLM-5.2 through confidential AI infra with Intel TDX + H100/H200, giving the model a privacy and attestation layer.
– @0G_labs: GLM-5.2 runs inside 0G Private Computer TEEs with signed attestation and on-chain verification.
– @VeniceAI | $VVV : already lists GLM-5.2 and an E2EE endpoint for private inference access.
– @chutes_ai | SN64 : serverless open-source model infra already serving GLM-5.2 around $1.40/M input and $4.40/M output.
– @c0mputeAI | $Zero : used GLM-5.2 as a live demo for Shard, serving a 744B model across prosumer GPUs over the public internet at ~30 tok/s.
– @virtuals_io | $Virtual : backed c0mpute after the GLM-5.2 distributed inference demo. c0mpute provides decentralized inference, Virtuals provides distribution through its agent ecosystem.
– @wardenprotocol | $Warden : fits the verified inference narrative with Halo and SPEX, though I haven't seen a confirmed GLM-5.2 integration yet.
GLM-5.2 gave #DePIN and crypto AI infra their best new excuse in months.
A huge open model creates immediate surface area for GPU markets, private inference, TEE receipts, token-metered APIs, agent wallets, and verifiable compute.
I see a few plays here.
Meat Stack (3/3) retweeted
🚨 CRAZY: OVER THE PAST FEW YEARS, BITCOIN WHALES BOUGHT BTC LIKE NEVER BEFORE.
Once Bitcoin bottoms...
The next bull market will be WILD. 👇
43% of Arena Cohort 0 inference spend went to traces with wrong outputs. Passing trajectories scaled reasoning by 10-35% depending on task difficulty. Ranks 7-15 fell sharply behind the top six on harder evaluations. Full breakdown at @SentientAGI.































