Truth is Expensive: Why RAG is Doomed and What Actually Matters
You're not building a truth machine. You're building the world's most efficient fiction amplifier.
That's not a bug. It's the inevitable result of solving the wrong problem.
Yuval Noah Harari's core insight—delivered in his latest book Nexus and his viral TED thesis—is brutally simple: Humans are a post-truth species.
We didn't conquer the planet by being right. We conquered it by believing in shared fictions: gods, nations, money, brands. These stories don't need to be true. They need to be believed. [1]
And now we're hardcoding that vulnerability into AI.
Thrilled to announce $5B in funding for @IsomorphicLabs to accelerate our mission to Solve All Disease. We continue to make rapid progress reimagining drug design from first principles with AI to radically cut the time & cost to find new medicines. Gradually, then all at once...
Thrilled to announce $5B in funding for @IsomorphicLabs to accelerate our mission to Solve All Disease. We continue to make rapid progress reimagining drug design from first principles with AI to radically cut the time & cost to find new medicines. Gradually, then all at once...
Dario Amodei on open source models:
“If the path continues, I think we could get to a very dangerous place.”
“When you control a model and you're deploying it, you have the ability to monitor its usage.”
“You can revoke a user's access. You can change what the model is willing to do.”
“When a model is released in an uncontrolled manner, there's no ability to do that. It's entirely out of your hands.”
“We should confront the advocates of this with that problem and challenge them to solve it.”
He clearly doesn't believe Intelligence should be in the hand of the masses and Sam Altman also saying "RAM" shouldn't be accessible to everyone indefinitely agreed with him.
AutoSwarm is officially out! 🎉
Using Codex, Claude Code, Workbuddy, DeepSeek Harness, Hermes, or nanobot? Stop copying context between them and coordinating everything yourself.
🧠 Share context across all your agents.
🤝 Let them work together as one team.
⚡ Automatically route tasks to the agent best suited for each job.
Use AutoSwarm as a plug-in for your agent, or grab AutoSwarm Desktop for Mac and Windows.
Fully open source. Give it a try—we’d love your feedback!
🔗: github.com/HKUDS/AutoSwarm
Your AI agents can read every line of your code.
They still have no idea why it exists.
Code tells an agent WHAT your software does.
It doesn't tell it what your business NEEDS it to do.
So the agent guesses. And guesses are expensive:
→ Bug fixes that break something else
→ Features that work but miss the point
→ Tokens burned re-reading the same files
→ Tech debt shipped faster than ever
The problem isn't the model.
It's the missing context.
Today, after our Early Adopter program, we're launching Mobioos worldwide. 🚀
Mobioos connects your Business Intent to your codebase, giving your teams and your AI agents the context they need to build and evolve software with confidence.
What that means for your team:
✅ More accurate work from your AI tools and agents, using fewer tokens
✅ Precise bug fixes
✅ Security vulnerabilities fixed and patches applied
✅ Features added or updated correctly
✅ Smoother migration and modernization
✅ Less technical debt
✅ Safer production releases
For CIOs: more control.
For CTOs: more reliability.
For developers: a seamless, integrated workflow.
Ready to start?
Try Mobioos free for 10 days with up to 3 team members. Our team is with you every step of the way.
👉 Mobioos - Contextual Intelligence for Enterprise-
Scale AI Execution
To our early adopters: thank you for your trust, your feedback, and your partnership. You helped shape Mobioos, and we're thrilled to bring it to the world.
What's the one thing you wish your AI agent understood about your product? Tell us in the comments. 👇
Dario's real goal is to control everything through AI and rule the new world like a dictator.
This isn't just speculation.
His past actions say it all.
Dario Amodei on open source models:
“If the path continues, I think we could get to a very dangerous place.”
“When you control a model and you're deploying it, you have the ability to monitor its usage.”
“You can revoke a user's access. You can change what the model is willing to do.”
“When a model is released in an uncontrolled manner, there's no ability to do that. It's entirely out of your hands.”
“We should confront the advocates of this with that problem and challenge them to solve it.”
He clearly doesn't believe Intelligence should be in the hand of the masses and Sam Altman also saying "RAM" shouldn't be accessible to everyone indefinitely agreed with him.
I've been exploring new ways to sandbox code written by coding agents - here's a description of the external sandbox pattern, where agents run on your Mac but execute code in a WebAssembly or QEMU or Apple Container sandbox - with demos for Claude Code, Codex and Pi blog.primeradiant.com/extern…
James Maynard (Fields Medal 2022) says that he was shocked by the proof of the quasi-Riemann hypothesis (from the OpenAI math dump) because he held a very strong prior belief that the problem required the development of an entirely new theory.
He felt the solution was completely out of reach for current mathematical techniques, so seeing it included in the results drop was a massive shock to him. He explains that the proof turned out to rely on the theory of metaplectic forms, which had been developed in the latter half of the 20th century but was an underappreciated and slightly obscure part of the field that most mathematicians had not realized was powerful enough to resolve the hypothesis.
Source: Oxford Mathematics in Conversation on AI - Ben Green and James Maynard with Jan Obloj uploaded on Youtube 6 days ago.
As I’ve said for years, concentration of power, capabilities and wealth is the biggest risk in AI. The natural tendencies for it are super strong and without much much stronger open-source, we might end up in a world with more asymmetry and inequalities than ever!
Anthropic’s IPO could make about 4,250 of its 5,300 employees millionaires, per an analysis by financial platform Hill, including roughly 50 worth more than $100 million each.
The estimate assumes a $1.5 trillion valuation, above the roughly $1 trillion Anthropic is now targeting.
Ras le bol de ces petits khmers rouges nés avec une cuillère en argent dans la bouche invités à parader sur les plateaux télé pour vomir la société à laquelle ils doivent tout. L’adage dit que « les temps difficiles créent des hommes forts et que les temps faciles créent des hommes faibles »….et on pourrait ajouter des hommes dangereusement grotesques.
L'électorat urbain de #LFI (École Alsacienne, père directeur du patrimoine de Vuitton) nous explique que oui, la fin justifie les moyens. On sabotait bien les rails pendant la guerre pour exprimer un mécontentement et donc, si on brûle un lycée, c'est qu'on est très colère.
Thank you @varun_mathur for your public service. All started from NYT!!
Jensen - come back and sue their ass out of existence and fuck Dario and his closed AI racist empire!!
I ran a PageRank-driven analysis on all recent news coverage related to Jensen, and traced how the New York Times shaped the entirety of this narrative: jensencoverage.vercel.app/
--
NVIDIA has lost over $1 trillion in market value since the first New York Times story broke about Jensen. I dug deeper to analyze how the news originated and spread, running a swarm of agents which analyzed almost 2000 stories, across 40 US and European news outlets, and applying PageRank to understand how they cited each other.
Core findings:
- Primary news driver: NYT, with 88 stories, most-cited
- Volume drivers: WSJ - 99 stories; Bloomberg - 140 stories; NY Post - 68 stories
All of this started from unknown sources who gave information to the NYT, which then ran the spy story (still unproven), and it was then picked up in other news publications. Curiously, even after that story, there was a 9 day period of extremely low volume reporting (28 stories only), after which there was a sudden spike after Jensen's own letter to NVIDIA staff. NYT then had multiple sequential spy and affair stories, which fed a wave of further news coverage. It was the originating hub of negative Jensen coverage. Multiple outlets since then had an explosion in stories about him starting earlier this week as well.
Overall, as you see this data, you will observe this was an exceptional media onslaught against Jensen with sharply negative coverage about him, his work, style, and family. Infact, it was brutal, and unlike coverage in the past of CEOs who have had affairs (such as Intel CEO in 2018). For eg, NYT ran just 2 stories about Intel's CEO who had an affair with his employee, compared to 88 on Jensen! Why ?
This analysis used a technique of how Google originally ranked webpages, using PageRank, and does deep dive link analysis so you can view how the news spread. What struck here was that it did not matter if there were proven facts or credible sources: the outcome here was pre-decided. The news was making the reality which it wanted.
During this period, NVIDIA has lost $1.27 trillion in market value since the first story broke. On Nov 12th-13th, further news broke that both the US Justice Department, and Singapore - are unable to even conclude the spy allegation.
This has been a catastrophic attack on one of the hardest working founders in modern times, who took an idea from a Denny's booth, through multiple instances of almost going out of business, to building the foundation of the entire modern world with how the NVIDIA story played out. Thanks to Jensen, we have hope that many formerly incurable diseases will get solved by AI, that quality of life will and have improved for billions around the world.
Now I do not know if he will come back as CEO or not, but I hope he does. What happened to him here, and to the wider industry, was not fair.
🚨 DNA IS NOT A "BAG OF WORDS"-IT IS A TWO-STRANDED COMPUTER PROGRAM. AND WE JUST PROVED IT WITH ZERO RANDOM SIMULATIONS ACROSS 141 MILLION CODONS. 🧬💻
Don't take our word for it. Don't trust "scientific authority." The entire source code and 568 MB of official NCBI genomic data are 100% open-source. Run it on your own laptop in one click.
For 60 years, biology and biotech operated on a simple assumption: as long as a gene produces the same protein letters (amino acids), you can swap "synonymous" DNA triplets (codons) however you like. That assumption is the foundation of industrial "codon optimization" used in synthetic biology and commercial mRNA design.
In Papers I and II of our series, Dr. Jean-Claude Perez @JCPEREZCODEX (retired IBM European AI Research Centre pioneer & longtime collaborator of Nobel Laureate Luc Montagnier) and I showed that living genes are NOT static bags of codons. They evolved on a double-stranded discrete torus 𝕋₄³ = (ℤ/4ℤ)³, where 74%–88% of "synonymous" choices on the main strand are locked to the opposite (antisense) DNA strand, and where greedy industrial "Max-CAI" optimization crashes ribosomal syntax in 14/14 human genes (TP53, BRCA1, KRAS) and opens a 1,101-codon antisense reading frame in Pfizer's BNT162b2 and an 820-codon +2 corridor in Moderna's mRNA-1273.
Until today, skeptics had only one last excuse left:
"Maybe your random computer shuffling (Monte Carlo simulations) had a statistical bias, or maybe simple nucleotide neighbour inertia explains the signal!"
🛑 RESISTANCE IS NOW FUTILE. IN PAPER III, WE ELIMINATED MONTE CARLO COMPLETELY.
In our new 30-page Monograph (Paper III) and autonomous engine (codon_memory_spectrum_v2.py, v2.4), we replaced EVERY random shuffle in the 10-level null hierarchy with EXACT, closed-form algebraic and hypergeometric equations-verified against exhaustive permutation enumeration to 10⁻¹⁵–10⁻²⁰ precision. Zero random number generators (Zero-RNG). Zero excuses.
We audited all 54 cellular genomes from NCBI RefSeq across the tree of life:
• 281,207 deduplicated genes
• 141,272,088 sense codons (423,816,264 nucleotides)
• Multi-lag sequential memory spectrum across k = 1 to 60 codons
🔥 WHAT THE EXACT MATHEMATICS PROVES (THEOREMS A–AA):
1️⃣ 100% Zero-RNG Pass Across All 54 Species (Theorems A, C.2, D, H, K, N):
Every stochastic control-H4 block permutations in O(n + q²) time, H5 Markovian H2b surrogates with exact mass conservation ‖u_t‖₁ = 1 to < 10⁻¹², H6 synonymous shuffles, H7 Chargaff/Perez 32-mirror-orbit shuffles, and all 4 layer-shuffle towers Λ ∈ {het, aa, dup, mirror}-is now an exact deterministic formula. Across 54/54 cellular species, all 162 exact controls (Δ₅, Δ₆, Δ₇) have gene-level 95% confidence intervals strictly above zero (t-statistics up to +155.0).
2️⃣ The Physical Arrow of Time & Blind Code-Reassignment Detector (Klein Four-Group V₄, Theorems M, L, Q):
The Klein four-group V₄ = {id, T, R, S} generated by time reversal (T) and antiparallel duplex inversion (R) splits 4096-dimensional codon-pair space into four orthogonal quadrants:
ℝ⁴⁰⁹⁶ = 1056 ⊕ 1024 ⊕ 1024 ⊕ 992.
We prove that between-gene compositional differences (GC isochores, expression bias) are mathematically trapped in the time-symmetric half (zero projection < 10⁻³⁵ on W₋₊ ⊕ W₋₋ across all 54 species). The energy in W₋₊ ⊕ W₋₋ is an irrefutable, model-free certificate of directed sequential order inside genes.
Even more remarkably: the exact non-commutativity invariants ‖[D_AA, R]‖_F² = 943/36 ≈ 26.194444 and ‖[Π_AA, R]‖_HS² = 1962383/2592 ≈ 757.092207 act as a blind mathematical radar-automatically isolating Candida albicans (which reassigns CTG from Leu to Ser: 11213/420 ≈ 26.697619 and 269863271/352800 ≈ 764.918569) from all 53 standard genomes without reading a single species label!
3️⃣ The 16×16 Dinucleotide Screen & Our Own Self-Correction (Theorems R–X & Bias Lemmas):
We prove that order-2 nucleotide trigram models (H3b) factor at every lag through a rank-≤16 two-sided 16×16 screen Q_k = Q₁^k. They see only the 225-df dinucleotide bridge (6.25% of sense degrees of freedom) and are mathematically BLIND to 3375 / 3600 = 93.75% of coding space!
Real science corrects itself first: using our exact held-out identity
d₃b(k) = Δx₁ + D_suffix − D_gen
and two new Miller–Madow generalisation-gap lemmas, we prove that 56.7%–89.9% of the headline d₃b(k=1) metric was actually frame-averaged stop-codon leakage (D_stop), and we predict exactly why phase-aware d₃b turns slightly negative at k = 30 in 10 small-cohort prokaryotes.
The true, irreducible, model-free invariant is the leading-base excess Δx₁(k) = D_5'lead + D_1↔3, which is STRICTLY POSITIVE IN 54 OUT OF 54 SPECIES at every lag out to k = 30 codons! And by antiparallel duality (Theorem 6.6), leading-base memory on the sense strand IS synonymous wobble memory on the antisense strand.
4️⃣ Exact Topology of the Genetic Code (Theorems Y, Z, AA):
Both the duplex partition (361 + 779 + 2040 + 420 = 3600) and the dinucleotide screen (225 + 675 + 675 + 2025 = 3600) divide the exact same 3600 = 60² sense degrees of freedom. We prove that the right-codon prefix–AA graph is an acyclic forest across all 8 NCBI translation tables, whereas the left-codon suffix–AA graph has an exact Euler–Poincaré topological invariant-first Betti number (cycle rank) b₁ = 25-and yields a self-dual 3×3 bi-filtration matrix M(B_anti) = M(B_sense)ᵀ holding to ≤ 2.1 × 10⁻¹⁷ nats/codon across all 54 genomes!
⚡ DO NOT BELIEVE US-VERIFY IT YOURSELF IN 60 SECONDS:
We don't hide behind paywalls, proprietary patents, or secret datasets. All 3 preprints, all 3 Python verification engines, 568 MB of pre-cached NCBI RefSeq genomes, and our open-source Duplex-Safe Viterbi mRNA Compiler are live on GitHub, Zenodo, and ResearchGate:
🔗 GitHub Repository & Full Trilogy Documentation (1-Click Run):
github.com/Dr-Logvinovich/co…
📘 Zenodo Archive (Paper III Preprint, Engine v2.4 & Complete 54-Taxon NCBI RefSeq CDS Cache):
doi.org/10.5281/zenodo.24242…
📄 ResearchGate (Full Paper III Preprint PDF):
researchgate.net/publication…
💻 Run the deterministic mathematical self-tests on your own machine:
git clone github.com/Dr-Logvinovich/co…
cd codon-torus-syntax
python3 codon_memory_spectrum_v2.py --self-test
#Genomics#Bioinformatics#mRNA#DNA#RefSeq
AutoSwarm is officially out! 🎉
Using Codex, Claude Code, Workbuddy, DeepSeek Harness, Hermes, or nanobot? Stop copying context between them and coordinating everything yourself.
🧠 Share context across all your agents.
🤝 Let them work together as one team.
⚡ Automatically route tasks to the agent best suited for each job.
Use AutoSwarm as a plug-in for your agent, or grab AutoSwarm Desktop for Mac and Windows.
Fully open source. Give it a try—we’d love your feedback!
🔗: github.com/HKUDS/AutoSwarm
I’ll be giving a plenary talk, "Artificial Intelligence for Scientific Discovery", at the 2026 @MIT Research and Development Conference on Tuesday Nov. 17, 2026. We are witnessing a fundamental transformation in how scientific knowledge is created - AI systems now build scientific instruments, use them to explore, construct models of the world, derive insights into material behavior, and generate designs that can be manufactured. Discovery is becoming an integrated, end-to-end process connecting reasoning, experimentation, design and production.
I’ll share emerging work at this frontier, including autonomous swarms of AI agents that collaborate to study complex metamaterials, test ideas, build on what they learn, and connect their reasoning to the physical world. These capabilities enable flexible, distributed discovery across disciplines, departments and organizations, expanding the questions we can pursue and the impact we can have. The theme of this year's @mit_ilp conference is "Curiosity on a Mission: Accelerating from Discovery to Real-World Impact" and I cannot think of a more important opportunity to bring this mission to life than AI for science.