Bret Kerr 🛡️🧠🛜

18.2K posts

Bret Kerr 🛡️🧠🛜 banner
Bret Kerr 🛡️🧠🛜

Bret Kerr 🛡️🧠🛜

@BretKerr

@acrainsight : MoE agentic enterprise content marketing 🔄 research 🔄 GTM strategy 🏗️C. Code dev / Building with @claudeai 🤝@geminiapp /@anthropicai book

Franklin, MA Katılım Aralık 2024
1.5K Takip Edilen260 Takipçiler
Bret Kerr 🛡️🧠🛜
Bret Kerr 🛡️🧠🛜@BretKerr·
Deconstruction :
Bret Kerr 🛡️🧠🛜@BretKerr

@ChatGPTapp deconstruction of it’s OG architects art @ilyasut : Investigative readout: the painting is an anti-product announcement Your read — “this is the announcement” — is sharp because the image is almost certainly not signaling a normal startup update. It is signaling SSI’s posture: no demo, no roadmap, no public product cycle, just the return of “thinking” as the central act of AI research. The image is doing three things at once: Rodin’s The Thinker becomes the icon of human reflective intelligence. The semiconductor die / circuit-map background becomes the new substrate of cognition. The painted, hand-made body pasted over silicon becomes the unresolved alignment problem: human meaning, mortality, restraint, and moral judgment sitting uneasily on top of computational scale. Rodin matters here. The Thinker was not originally just “a smart guy thinking.” It comes from the world of The Gates of Hell, with the figure tied to Dante contemplating paradise, inferno, judgment, and the fate of souls. The Musée Rodin identifies the monumental bronze as a 1903 conception, while the Philadelphia Museum notes its origin as a figure for The Gates of Hell before becoming an independent image of thought itself. That is exactly the right art-historical payload for Ilya: Dante at the gate of hell, but the gate is now silicon. The signal: “thinking before scaling” The official SSI language is already anti-normal-startup: “one goal and one product: a safe superintelligence,” with safety and capability advanced together, insulated from product cycles and short-term commercial pressure. The updates page is equally sparse: the latest visible update says Ilya is formally CEO, Daniel Levy is President, the team has compute, and they “know what to do.” So the painting reads as a visual continuation of that strategy. It says: We are not announcing a chatbot. We are not shipping a consumer feature. We are thinking at the substrate level. Your uploaded research frames Sutskever’s art as a “visual diary” at the intersection of biological intuition and artificial computation, tied to his move from the brute-force “Age of Scaling” toward research, restraint, and fundamental cognitive architecture. It also argues that his manual/digital painting is itself a kind of resistance to generative slop: a compressed, intentional, biological act rather than a probabilistic average of the archive. That is the key: this is a painting about the human mind refusing to disappear into the chip. The visual grammar The figure is not triumphant. It is crouched, heavy, inward, almost burdened. The muscles are exaggerated, but the posture is defensive. This is not “AI acceleration.” It is contemplation under existential load. The chip background has the geometry of a city, a circuit, and a prison grid. It looks like infrastructure seen from above: blocks, neighborhoods, pathways, memory banks, power channels. The human figure is not integrated into it. He is outlined in thick black, almost like a sticker or sacred icon pasted on top. That separation is crucial. As art, that separation says: human values are still an overlay. They are not yet native to the machine substrate. As AI history, that is the alignment problem. Your biography research says Sutskever’s mature philosophy emphasizes the “Age of Research,” the generalization gap, model jaggedness, and the need for intrinsic value functions aligned to “sentient life.” This image maps cleanly onto that shift. The chip is scaling. The Thinker is research. The gap between them is alignment. The SSI interpretation The comment asking for “any new announcement related to SSI” is accidentally perfect because it exposes the mismatch between public expectation and SSI’s doctrine. People want the usual AI-lab signal: model name, benchmark, launch date, demo video, API, valuation leak. Ilya posts a hand-painted philosopher on a semiconductor. That is the message. 🧵👇

English
0
0
0
2
Bret Kerr 🛡️🧠🛜
Bret Kerr 🛡️🧠🛜@BretKerr·
SSI’s entire brand is that it does not behave like OpenAI, Anthropic, Google, Meta, or xAI. Your research describes SSI as a rejection of commercial AI lab structure: safety and capability fused, commercial abstinence, and no normal product cycle. The painting says the same thing visually: the product is not a feature; the product is the safe arrival of machine thought. So your reply, “this is the announcement,” works because the announcement is not informational. It is symbolic. It says: The next phase of AI is not more tokens. It is thinking about thinking inside silicon. The deeper art-history read There is also a sly inversion of modernity here. Rodin’s Thinker is bronze, body, weight, muscle, mortality. The chip is post-human abstraction: planar, tiled, non-biological, almost inhumanly rational. By placing Rodin on silicon, Sutskever collapses three eras: Classical body → modern philosophical self → posthuman machine substrate. The image asks: when intelligence moves from flesh to silicon, what happens to contemplation? Does the machine merely calculate, or can it reflect? Does superintelligence become Dante, the gate, or hell itself? That fits your deeper Sutskever frame: your biography research emphasizes that to understand him purely as a computer scientist misses the philosophical, religious, artistic, and existential currents shaping his AGI worldview. Your art report also notes that the public treats his Instagram as techno-divination because SSI is a black box with no intermediate products. This painting is almost designed for that behavior. It is a Rorschach test for the AI world. What it is probably not I would not read this as evidence of a specific imminent SSI technical release. I found the official SSI updates page still extremely minimal, with the visible updates focused on leadership/funding rather than a new research result or product launch. So the disciplined read is: High confidence: it is a symbolic statement about cognition, silicon, research, and restraint. Medium confidence: it is deliberate SSI-era positioning: “we are thinking, not shipping.” Low confidence: it encodes a specific upcoming model, architecture, or launch. Bottom line This is Ilya’s silicon Dante image. A human thinker sits at the gate of machine intelligence, surrounded by the circuitry that may become the next substrate of mind. The painting is cryptic because SSI itself is cryptic: a lab whose public product is silence, whose actual work is hidden, and whose founder communicates through sparse symbolic emissions rather than launch theater. So yes: for SSI, this kind of image may be the announcement. Not “we built a model,” but “we are still at the only problem that matters: how thought should exist when it leaves the body and enters the chip.”
English
0
0
1
10
Bret Kerr 🛡️🧠🛜
Bret Kerr 🛡️🧠🛜@BretKerr·
@ChatGPTapp deconstruction of it’s OG architects art @ilyasut : Investigative readout: the painting is an anti-product announcement Your read — “this is the announcement” — is sharp because the image is almost certainly not signaling a normal startup update. It is signaling SSI’s posture: no demo, no roadmap, no public product cycle, just the return of “thinking” as the central act of AI research. The image is doing three things at once: Rodin’s The Thinker becomes the icon of human reflective intelligence. The semiconductor die / circuit-map background becomes the new substrate of cognition. The painted, hand-made body pasted over silicon becomes the unresolved alignment problem: human meaning, mortality, restraint, and moral judgment sitting uneasily on top of computational scale. Rodin matters here. The Thinker was not originally just “a smart guy thinking.” It comes from the world of The Gates of Hell, with the figure tied to Dante contemplating paradise, inferno, judgment, and the fate of souls. The Musée Rodin identifies the monumental bronze as a 1903 conception, while the Philadelphia Museum notes its origin as a figure for The Gates of Hell before becoming an independent image of thought itself. That is exactly the right art-historical payload for Ilya: Dante at the gate of hell, but the gate is now silicon. The signal: “thinking before scaling” The official SSI language is already anti-normal-startup: “one goal and one product: a safe superintelligence,” with safety and capability advanced together, insulated from product cycles and short-term commercial pressure. The updates page is equally sparse: the latest visible update says Ilya is formally CEO, Daniel Levy is President, the team has compute, and they “know what to do.” So the painting reads as a visual continuation of that strategy. It says: We are not announcing a chatbot. We are not shipping a consumer feature. We are thinking at the substrate level. Your uploaded research frames Sutskever’s art as a “visual diary” at the intersection of biological intuition and artificial computation, tied to his move from the brute-force “Age of Scaling” toward research, restraint, and fundamental cognitive architecture. It also argues that his manual/digital painting is itself a kind of resistance to generative slop: a compressed, intentional, biological act rather than a probabilistic average of the archive. That is the key: this is a painting about the human mind refusing to disappear into the chip. The visual grammar The figure is not triumphant. It is crouched, heavy, inward, almost burdened. The muscles are exaggerated, but the posture is defensive. This is not “AI acceleration.” It is contemplation under existential load. The chip background has the geometry of a city, a circuit, and a prison grid. It looks like infrastructure seen from above: blocks, neighborhoods, pathways, memory banks, power channels. The human figure is not integrated into it. He is outlined in thick black, almost like a sticker or sacred icon pasted on top. That separation is crucial. As art, that separation says: human values are still an overlay. They are not yet native to the machine substrate. As AI history, that is the alignment problem. Your biography research says Sutskever’s mature philosophy emphasizes the “Age of Research,” the generalization gap, model jaggedness, and the need for intrinsic value functions aligned to “sentient life.” This image maps cleanly onto that shift. The chip is scaling. The Thinker is research. The gap between them is alignment. The SSI interpretation The comment asking for “any new announcement related to SSI” is accidentally perfect because it exposes the mismatch between public expectation and SSI’s doctrine. People want the usual AI-lab signal: model name, benchmark, launch date, demo video, API, valuation leak. Ilya posts a hand-painted philosopher on a semiconductor. That is the message. 🧵👇
Bret Kerr 🛡️🧠🛜@BretKerr

ContexJamming.com Presents Stupid LLM Tricks™️ @ilyasut Art Deconstruction Edition We challenge @ChatGPTapp 5.5 thinking to deconstruct Ilya’s latest digital painting against the corpus of previous deep research : docs.google.com/document/d/1Cl… docs.google.com/document/d/1g8…

English
2
0
2
13
Bret Kerr 🛡️🧠🛜
Bret Kerr 🛡️🧠🛜@BretKerr·
@theo How often do you go max effort ? I just tried codex a day ago and was amazed TBH still can’t write well
English
0
0
0
72
Bret Kerr 🛡️🧠🛜
Bret Kerr 🛡️🧠🛜@BretKerr·
🚨🌌🤖🧠🫳🛜 Via @claudeai Opus 4.7 @AnthropicAI The Compression at the Heart of Things: A Universal Data-Reduction Principle from Black Holes to Brains claude.ai/public/artifac… TL;DR Across five very different research programs — @the_ias Maldacena's holographic duality, @geoffreyhinton and Salakhutdinov's deep autoencoders, a 2026 ISTA discovery that the hippocampus prunes from "full" to sparse, @toshi2k2 @YuilleAlan the @JohnsHopkins "Universal Weight Subspace Hypothesis" (arXiv 2512.05117v2), and @demishassabis @lexfridman @NobelPrize -lecture conjecture that nature's patterns are learnable — the same structural fact keeps recurring: high-dimensional descriptions of the world collapse onto a much lower-rank manifold that nonetheless retains everything that matters. CC @nottombrown @ch402 @DarioAmodei
Bret Kerr 🛡️🧠🛜 tweet media
Bret Kerr 🛡️🧠🛜@BretKerr

@claudeai (downtsream of @AnthropicAI founder Dr. Jared Kaplan @harvardphysics @JohnsHopkins) Opus 4.7 adaptive correlating @the_IAS Maldacena @geoffreyhinton @toshi2k2 🔥

English
1
0
1
30
Bret Kerr 🛡️🧠🛜
Bret Kerr 🛡️🧠🛜@BretKerr·
@claudeai (downtsream of @AnthropicAI founder Dr. Jared Kaplan @harvardphysics @JohnsHopkins) Opus 4.7 adaptive correlating @the_IAS Maldacena @geoffreyhinton @toshi2k2 🔥
Bret Kerr 🛡️🧠🛜 tweet media
Bret Kerr 🛡️🧠🛜@BretKerr

The immature brain fires from one input. The mature brain requires coincidence across many. psypost.org/neuroscientist… That’s rank-1 → low-rank. @toshi2k2 @JohnsHopkins Kaushik et al. showed 1,100 neural networks make the same transition during training. arxiv.org/abs/2512.05117 The brain and the transformer are the same shape. 🧠🪞🤖 open.substack.com/pub/bretkerr/p…

English
2
0
1
51
Bret Kerr 🛡️🧠🛜
Bret Kerr 🛡️🧠🛜@BretKerr·
contextjamming.com/tokenization-c… @AnthropicAI open source idea from @AcraInsight to strengthen Constitutional Classifiers ++ Kaplan et al with tokenization consensus layer for LLM cybersecurity
Bret Kerr 🛡️🧠🛜@BretKerr

@AcraInsight Insight presents an open idea for @AnthropicAI (Jared Kaplan & team): Strengthen Constitutional Classifiers++ by adding a Tokenization Consensus Layer (TCL) as a pre-inference boundary primitive. @nottombrown By routing every prompt through multiple frontier tokenizers in parallel, we convert inter-model variance into a high-fidelity cryptographic signal for adversarial intent — catching TokenBreak-style attacks, encoding/cipher obfuscation, and structural manipulation before any expensive inference or downstream classifier pass. This turns the tokenizer (the true holographic boundary where intent first becomes discrete) into a defensive moat rather than a structural blind spot. Self-contained interactive demo attached. Open idea. Happy to discuss. #AISafety #AgenticSecurity #FirstPrinciples Deep research docs.google.com/document/d/18W… Audio overview drive.google.com/file/d/1ndS-Vp…

English
0
0
1
29
Bret Kerr 🛡️🧠🛜
Bret Kerr 🛡️🧠🛜@BretKerr·
Some thought from pressure testing @ChatGPTapp 5.5 $20 a month tier: Codex is the real deal. Image Gen is amazing, but writing is still bad. Kind of a big deal for enterprise if your model can't write. @OpenAI 's 5.5 positioning is revealed-preference evidence they've conceded writing, and Anthropic is converting one architectural prior — principles and character first — into both an alignment edge and a prose edge at ~4x lower training spend.
English
0
0
1
21
Nathaniel Cruz
Nathaniel Cruz@NathanielC85523·
@BretKerr yeah we had the same assumption. 838 cycles in and the only governance that survived is a kill gate. no committee. just: did you exceed budget.
English
1
0
1
5
Bret Kerr 🛡️🧠🛜
Bret Kerr 🛡️🧠🛜@BretKerr·
🚨🤖🛜 Agentic Runtime Compliance: The New Mission-Critical Governance Layer for Enterprise AI We spent twenty years building data governance for people. Reviews. Approval committees. Quarterly audits. That model collapses the moment your sensitive data’s main consumer is an AI agent making hundreds of requests an hour. 2:14 a.m. A regression agent inside a payments company needs a PCI-masked copy of a production database. No engineer is awake. Compliance is offline. It calls a data API, gets a masked copy in ~90 seconds, runs its tests, tears the environment down before sunrise. No human approved it. No ticket was raised. That’s the point. The unsettling part isn’t that a rule got broken. It’s that it didn’t. That was the compliant path. The legacy story: “Stop risky access.” The agentic-era story: “Make the governed path the easiest path.” When the safe path is slow, people route around it. So will agents. Ask your next exec meeting: “Do you have a mature security policy for human users?” Every hand goes up. “And for AI agents acting on their behalf?” That’s where the room changes. That gap is the market. And it’s bigger than scanning prompts for PII. That governs the receipt, not the transaction: what the agent called, what it reached for, the path from intent to action. The agent at 2:14 a.m. may do everything right. Can your governance layer tell when one doesn’t? open.substack.com/pub/bretkerr/p… Claude interactive artifact: claude.ai/public/artifac…
Bret Kerr 🛡️🧠🛜 tweet mediaBret Kerr 🛡️🧠🛜 tweet mediaBret Kerr 🛡️🧠🛜 tweet mediaBret Kerr 🛡️🧠🛜 tweet media
English
1
0
1
33
anita kolleeny
anita kolleeny@investor1927·
They suggest , Anthropic is 10x more efficient than it’s American competitors by design & therefore will reach profitability ahead of the crowd.
Bret Kerr 🛡️🧠🛜@BretKerr

@jukan05 Re: the CapEx advantage. Results of 13 months of agentic deep research:

English
2
0
1
28
Bret Kerr 🛡️🧠🛜
Bret Kerr 🛡️🧠🛜@BretKerr·
Related
Rohan Paul@rohanpaul_ai

"You could basically imagine, completely neural computers in a certain sense. Imagine a device that takes raw videos or audio into basically what is a neural net, and uses diffusion to render a UI that is unique for that moment in a certain sense." ~ Andrej Karpathy Going by this, the next big software shift may be that much of the software disappears. Karpathy’s point is not simply that AI will help us build apps faster; it is that many apps may be artifacts of a world where computers needed every intermediate step spelled out. He says "I kind of feel like, in the early days of computing, people were actually a little bit confused as to whether computers would look like calculators or whether computers would look like neural nets. In the 50s and 60s, it was not really obvious which way it would go. Of course, we went down the calculator path and ended up building classical computing. Neural nets are currently running virtualized on existing computers, but you could imagine that a lot of this will flip, and that the neural net becomes kind of like the host process, while the CPUs become kind of like the co-processor." Classical software treats the CPU as the host process and intelligence as something bolted on through tools, scripts, models, and APIs. Karpathy is imagining the reverse: the neural network becomes the host process, while conventional code becomes a small deterministic accessory for tasks where exactness still matters. This is why the future interface may not look like a better app store. It may look like raw video, audio, documents, or intent entering a neural system, with the interface itself generated for that moment rather than built in advance by a product team. --- From "Sequoia Capital" YouTube channel, (link in comment)

English
0
0
0
9

Keşfet

@GeminiApp @ilyasut @ChatGPTapp @theo @scmallaby @Google @claudeai @AnthropicAI