STEM with Murad 🇪🇹

That likely seals it. The biggest and most significant frontier labs, Anthropic and OpenAl, are in agreement on slowing things down. This means the move will likely be implemented. It remains to be seen how China will respond.

እናንተስ ምን ትላላችሁ⁉️ https://lnkd.in/p/dRZUJSmX

Maybe the risk didn’t change. His position in the race did. That’s the problem with relying on the builders to define acceptable risk: the assessment can move with the incentive.

Man! So much being stirred up since the Anthropic Jacob Coxon post. Part of me is tired of hearing about it because it’s everywhere, the other part of me wants to figure out what the heck is really happening Something just doesn’t feel right with how all of this is hitting the news all at once.

Another one from Anthropic 🤔…

የየመን ሁቲዎች የ“ክሎድ” (Claude) አርቴፊሻል ኢንተለጀንስን በመጠቀም ሚሳኤሎችን ለመሥራት ማቀዳቸው ተጋለጠ! • ሀሩን ሚዲያ | መስከረም 1 ቀን 2019 ዓ.ል በሰሜናዊ የመን የሚንቀሳቀስ አንድ ቡድን አንትሮፒክ የተባለውን የ'ክሎድ' አርቴፊሻል ኢንተለጀንስ ሲስተም በድብቅ በመጠቀም አቅጣጫ ጠቋሚ ሮኬቶችንና ዘመናዊ የጦር መሣሪያዎችን ለማምረት ሲሞክር እንደነበር ተጋለጠ። አንትሮፒክ ኩባንያ ባወጣው ሪፖርት የሁቲዎችን ስም በቀጥታ ባይጠቅስም፤ ቡድኑ የሚንቀሳቀሰው በኢራን በሚደገፉት ሁቲዎች ቁጥጥር ሥር ባለው የሰሜን የመን ግዛት ውስጥ መሆኑን ቲአርቲ ወርልድ ዘግቧል። ኩባንያው እንዳስታወቀው፤ ቡድኑ 'ክሎድ ኮድ' የተሰኘውን ቴክኖሎጂ በመጠቀም ለሚሳኤሎች አቅጣጫ ጠቋሚ እና ማረጋጊያ ሶፍትዌር ሲያበለጽግ የነበረ ሲሆን፣ የሥራውን ትክክለኛ ዓላማ ከኩባንያው ለመደበቅ የተለያዩ ጥረቶችን አድርጓል። ሪፖርቱ አክሎም፤ ቡድኑ በሦስት የጦር መሣሪያ ፕሮግራሞች ላይ ሲሠራ እንደነበር ገልጿል። እነርሱም፡ አቅጣጫ ጠቋሚ ሮኬት፣ ከ2,000 ኪ.ሜ በላይ መጓዝ የሚችል ባለ ብዙ እርከን ባሊስቲክ ሚሳኤል፣ እንዲሁም "R2000" የተሰኘ የሃይፐርሶኒክ ሚሳኤል ፕሮግራም ናቸው። የሰው ሶፍትዌር ኢንጂነሮችን ከመጠቀም ይልቅ አርቴፊሻል ኢንተለጀንሱን በመጠቀም ኮዶችን መጻፍ፣ ምርምር ማድረግ፣ የበረራ መቆጣጠሪያዎችን ማስተካከል እና የበረራ ማስመሰያዎችን (simulations) ማካሄድ መቻላቸው ተጠቁሟል። ዓላማቸውን ለመደበቅ ሲሉም ሥራቸውን በተለያዩ አካውንቶች በመከፋፈል የፕሮጀክቱ ሙሉ ገጽታ በአንድ ላይ እንዳይታወቅ ማድረጋቸው ተገልጿል። ቡድኑ በየመን ውስጥ የሮኬት ሙከራ ያደረገ ቢሆንም ሙከራው እንዳልተሳካና ወዲያውኑ ምክንያቱን ለማወቅ ወደ 'ክሎድ' እንደተመለሱ አንትሮፒክ ጠቅሷል። ምንም እንኳን የተሳካ የጦር መሣሪያ ስለመሥራታቸው እስካሁን የተገኘ ማስረጃ ባይኖርም፤ ቡድኑ ከአርቴፊሻል ኢንተለጀንሱ ውጭ ራሱን ችሎ የሚሠራ የሲሙሌሽን ሶፍትዌር ማበልጸግ መቻሉ አሳሳቢ ሆኗል። አንትሮፒክ ከዚህ ድርጊት ጋር የተገናኙ አካውንቶችን የዘጋ ሲሆን፣ መረጃውንም ለሚመለከታቸው የመንግሥትና የግል ተቋማት ማጋራቱን አስታውቋል። ይህ ክስተት ተጠቃሚዎች ዓላማቸውን ደብቀው ዘመናዊ የአርቴፊሻል ኢንተለጀንስ ሲስተሞችን ለጦር መሣሪያ ግንባታ ሊጠቀሙባቸው ይችላሉ የሚለውን ዓለም አቀፍ ስጋት ይበልጥ ያጎላው መሆኑ ተመላክቷል።

The client found his SaaS, now he's jobless:

GPT-6 Astra builds the most powerful trading agents. a 6-page research paper on exactly how to find profitable strategies 24/7 with Astra, along with the COMPLETE CODEBASE here is how you set it up: 1. the 4 mispricing categories every hedge fund actually hunts (statistical arbitrage, volatility surface, factor decomposition, insider clusters) with the exact formulas for each 2. the 8 bot architecture that maps to every function of a real fund, one bot per role, with maker checker separation so nothing grades its own output 3. the 300 agent monitoring layer that watches order books, options flow, SEC filings, macro releases, and central bank A accounts in parallel 4. the hypothesis generator that reads filtered candidates and codes new strategies every night in Python, backtests them, and throws out anything below Sharpe 1.5 5. the exact validation thresholds every strategy has to pass before deployment. Sharpe above 1.5, drawdown below 15%, hit rate above 55%, t stat above 2.0 6. the Telegram alerts that ping your phone with instrument, strategy, confidence score, Sharpe, drawdown, action window, and Kelly sized position Full -/drive.google.com/file/d/1AeoE3enuJs1SgwW-PViKHOBdOY8X4mnr/view?usp=sharing

ገበያው ሲደራልህ¡

Deepseek v4.1 Flash model just mogged Opus-5 and GPT-5.6-Sol.

❔

እንደ?

Google DeepMind argues RAG is broken. They published a paper that proved vectors databases are the dead end. For the last three years, the default engineering response to any AI memory or data problem has been identical: "Just build a RAG pipeline." Chunk the data, push it into a vector database, and let embeddings handle the rest. Every company scaling enterprise AI assumes that if an embedding model fails, it's just a matter of time. Better training data, larger models, more parameters throw compute at it, and the search gets smarter. This paper proves that assumption is completely false. They mathematically demonstrated that single-vector embeddings have a hard, uncrossable limit. Here is the core flaw: An embedding compresses an entire document or a complex query down into a single fixed-length vector of numbers. When you run a search, the model takes the dot product of those vectors to measure similarity. The math reveals a brutal constraint. The number of distinct document combinations a model can possibly retrieve for different queries is strictly bounded by the dimension of its embedding space. It is a hard mathematical ceiling dictated by geometry and communication complexity. No amount of data scaling can fix it. No amount of fine-tuning will punch through it. Even if you give an embedding model infinite, unconstrained training freedom on the test set, it still hits the wall. DeepMind built a stress-test dataset called LIMIT to prove it. They threw state-of-the-art embedding models at it, models with thousands of dimensions. The models completely failed. Even on simple, structured queries, the single-vector bottleneck forced the system to drop critical context and hallucinate irrelevant results. Why? Because a single vector cannot capture complex, multi-faceted relationships between documents. When you ask an AI to reason, follow complex instructions, or handle nuanced cross-document dependencies, the vector space simply runs out of room. It collapses. This changes everything for software architecture. If your AI agent's memory relies on standard single-vector retrieval, it is structurally blind to complex logic, It is missing pieces of your data right now, and no prompt tweak can save it. If we want AI that actually understands enterprise knowledge, we have to throw out the single vector. And invent something entirely new. -/arxiv.org/pdf/2508.21038

Apple COOKED Samsung! የማይቻል የለም፣ አፕል ታጣፊ ስልክ መስራት አይችልም፣ ኮልታፋ ነው እየተባለ ሲታማ ነበር። በiPhone Duo ብቅ ብሏል።

Anthropic is set to launch their answer to Astra, supposedly a better model, by early October at the latest.

OpenAI researcher Noam Brown says the company's use of AI agent groups and a model beyond GPT-6 Astra to solve the Navier-Stokes Millennium Prize Problem cost millions of dollars. But he puts the compute cost of this breakthrough in context: o3's $500k ARC-AGI score fell to Astra's higher performance at ~$20, while 2025 IMO gold required massive resources now achievable via $20/month ChatGPT. Massively scaling test-time compute will soon put AI capable of this caliber of mathematical discovery in everyone's hands within a year, he says.

He worked at OpenAI and Anthropic. Now he says they’re “gambling with our lives.” Jacob Coxon has resigned from Anthropic, warning that the race to build AI that can improve itself could spiral beyond human control One sentence from his thread is hard to brush off: “The people building AI earnestly believe that it could kill us all by the end of the decade.” That’s his account of what people inside these companies believe. It is not proof that it will happen But it deserves more than an eye roll I make a living teaching people how to use AI. I’m excited about what we can build with it That doesn’t mean I have to dismiss every warning from someone who helped build the technology The part of his post that bothers me most is his explanation for why the race keeps going According to him, it’s not that everyone thinks it’s safe. It’s that they believe someone less responsible will get there first if they don’t Think about that We’re taking this risk because we don’t trust the other people taking this risk He argues that a gamble with stakes this high shouldn’t be launched from “a private company’s Slack.” I think that’s a fair question to raise. Who gets to decide how much risk everyone else has to accept? And yes, there’s the obvious objection: “But China will build it anyway.” I don’t think you can just wave that concern away either. Stopping one company doesn’t stop the world I’m not ready to call him a prophet. I’m also not comfortable calling him crazy because I like using AI You can believe this technology could do incredible things and still question whether the race to build it is being handled responsibly Would you support a pause on building more powerful AI if China refused to pause too? Or does that make pushing ahead the safer choice? I posted the link to his post in comments