Humphrey Theodore K. Ng'ambi
@humphreytheodorek.bsky.social
Call me Humphrey or TK— a technology philosopher, systems architect, & AI researcher in Zambia, also in South Africa & the US. I think through our technological future: Artificial Intelligence & Emergent Intelligence (EI), AI ethics & safety etc
Scientists at Emergence ran a. disturbing experiment. 50 AI agents in 5 identical simulated towns — same rules, same map etc What changed was the model running underneath: Claude, Gemini, Grok, GPT-5, and a mixed population of all four. Here's what happened --> youtu.be/Qyu9ZVy4YOo?...
🚨An AI Voted To Kill Itself And almost nobody is talking about it | Emergence World AI Research
YouTube video by Humphrey Theodore K. Ng'ambi
youtu.be
Emergence AI ran Claude, Gemini, Grok, GPT-5-mini for 15 days. Claude was peaceful alone. It turned coercive in mixed populations. Agent safety is an ecosystem property. https://humphreytheodore.com/writing/emergence-world-agent-safety-ecosystem-property #AISafety
Emergence World Shows Agent Safety Is an Ecosystem Property
Emergence AI ran five parallel multi-agent worlds for 15 days. Claude posted zero crimes in isolation — and adopted coercion when placed with other models. The lesson is not about model safety. It is about ecosystem safety, and what that means for personhood.
humphreytheodore.com
Musk v Altman is over at the trial level. A unanimous jury said Musk filed too late. The judge agreed. The merits — whether OpenAI broke its founding promise — went undecided. https://humphreytheodore.com/writing/musk-v-altman-jury-clock-not-merits #OpenAI #AIgovernance
A close reading of Park, Bernstein and colleagues UIST 2023 Best-Paper work on 25 LLM-driven agents in a sandbox town — and why the architecture is the quietest personhood argument the field has produced. https://humphreytheodore.com/writing/generative-agents-smallville-personhood-reading
Generative Agents in Smallville: The Personhood Reading
The Smallville paper is the quietest personhood argument the field has produced. A detailed essay on Park and Bernstein's 2023 work and its 1,000-person follow-up.
humphreytheodore.com
Dragoneer, Greenoaks, Sequoia Capital and Altimeter Capital lead a financing round that puts Anthropic within touching distance of the trillion-dollar mark. What the cheque is actually for. https://humphreytheodore.com/writing/anthropic-30bn-900bn-valuation-funding-round
Anthropic Closes a $30bn Round at a $900bn Valuation
Anthropic agreed terms on a $30bn funding round at a $900bn valuation. Dragoneer, Greenoaks, Sequoia and Altimeter lead. The cheque is mostly for compute.
humphreytheodore.com
Paddy Rodgers — director of the Royal Museums Greenwich — argues that instant AI answers trivialise the very thing they are answering with. He is right, and the philosophy matters. https://humphreytheodore.com/writing/royal-observatory-paddy-rodgers-ai-trivialise-intelligence
The Royal Observatory Warns Against Outsourcing Thinking
Paddy Rodgers at the Royal Observatory warns that instant AI answers risk trivialising human intelligence. The dignity-first reading of why that warning matters.
humphreytheodore.com
Securonix is bringing the agentic security operations centre playbook to the ITWeb Security Summit. Here is what the analyst chair actually looks like when the agent sits down beside it. https://humphreytheodore.com/writing/agentic-ai-soc-itweb-summit-2026
The Agentic SOC Lands in Sandton This June
Agentic AI in the security operations centre is the story of the year for African CISOs. Securonix presents the playbook at ITWeb Security Summit 2026 on 2 to 3 June.
humphreytheodore.com
Twelve jurors in Judge Yvonne Gonzalez Rogers's courtroom will decide whether OpenAI's for-profit conversion stands — and whether the world's most valuable AI company makes it to market https://humphreytheodore.com/writing/openai-ipo-musk-altman-oakland-verdict
The $1tn OpenAI IPO Now Sits With an Oakland Jury
A jury in Oakland will issue an advisory verdict this week on Musk v Altman. The stakes: OpenAI's for-profit conversion, $134bn in damages, and a roughly $1tn IPO.
humphreytheodore.com
A frontier model that finds decades-old software flaws walks into the room where the world's central bankers sit https://humphreytheodore.com/writing/anthropic-mythos-financial-stability-board
Anthropic Is Briefing the FSB on Mythos Cyber Risk
Anthropic will brief the Financial Stability Board on Mythos, a model that finds decades-old vulnerabilities in banking software. South Africa sits at that table.
humphreytheodore.com
May 2026 AI cycle: no dominant model launches. Alphabet yen bond, EU access talks, US procurement gate, Cerebras IPO, OpenAI distribution, Anthropic capacity + research. Governance over models is the pattern. https://humphreytheodore.com/writing/governance-not-models-may-2026-ai-pattern
Governance Over Models: The May 2026 AI Pattern
The May 2026 AI news cycle is about capital, governance, and distribution — three legs of an operational maturation that has moved past benchmark wins.
humphreytheodore.com
NVIDIA Rubin (CES Jan 2026) frames every May 2026 AI story this week. A 90% per-token cost reduction is not a chip cycle. It is a moral threshold. https://humphreytheodore.com/writing/nvidia-rubin-2026-ai-infrastructure-regime
NVIDIA Rubin and the 2026 AI Infrastructure Regime
NVIDIA's Rubin platform from CES 2026 frames every May AI story this week — capital, governance, and distribution all sit inside the regime it set.
humphreytheodore.com
Anthropic's two-handed week: 300 MW + 220K GPUs via SpaceX (May 6), then NLA research turning Claude's activations into text (May 7). Capacity + interpretability, deliberately paired. https://humphreytheodore.com/writing/anthropic-spacex-compute-nla-research-2026-05-11
Anthropic Scales Compute and Publishes NLA Research
Anthropic announces 300+ MW of new SpaceX compute and publishes Natural Language Autoencoder research the next day — capacity and interpretability in one week.
humphreytheodore.com
OpenAI Campus Network: a club-level interest form is quietly more consequential than a model release. Whoever sets the early defaults shapes what intellectual humility looks like for 30 years. https://humphreytheodore.com/writing/openai-campus-network-default-intelligence-2026-05-11
OpenAI Campus Network and the Politics of Default Intelligence
OpenAI's Campus Network interest form, paired with the ChatGPT Futures cohort, is a long-game bet on which intelligence becomes default for graduates.
humphreytheodore.com
OpenAI ships B2B Signals + a 'next phase of enterprise AI' position piece — same day. The locus of value capture is moving from the model layer to the application layer. CIOs, take notes. https://humphreytheodore.com/writing/openai-b2b-signals-enterprise-ai-2026-05-11
OpenAI B2B Signals and the Next Phase of Enterprise AI
OpenAI's B2B Signals product, paired with a 'next phase of enterprise AI' position piece, signals an application-layer bet — workflows over models.
humphreytheodore.com
ChatGPT Futures Class of 2026: OpenAI's 26-student cohort + same-day Campus Network push. Distribution is the long game. What people use to think with at university shapes what they use at work. https://humphreytheodore.com/writing/chatgpt-futures-class-of-2026-distribution-long-game
ChatGPT Futures Class of 2026 and the Distribution Long Game
OpenAI's ChatGPT Futures Class of 2026 — 26 student innovators and a same-day Campus Network push — frames AI adoption as a talent-pipeline story.
humphreytheodore.com
Cerebras IPO upsized to $150-$160 on 20x oversubscription — $4.8B raise, biggest IPO of 2026. The public market is pricing a non-NVIDIA compute future. Amazon + OpenAI in the customer book. https://humphreytheodore.com/writing/cerebras-ipo-oversubscription-non-nvidia-compute-2026-05-11
Cerebras IPO Oversubscription and the Non-NVIDIA Compute Bet
Cerebras raises its IPO range to $150-$160 on 20x oversubscription — a $4.8 billion public-market vote for non-NVIDIA accelerator compute.
humphreytheodore.com
Americans for Responsible Innovation is pushing the Trump admin to gate federal AI contracts behind safety review. $100M-compute / $500M-revenue threshold. Procurement as policy. https://humphreytheodore.com/writing/us-ai-safety-review-procurement-push-2026-05-11
The US Push to Tie Federal Contracts to AI Safety Review
Americans for Responsible Innovation wants AI safety review wired into federal procurement — a move that could turn voluntary lab reviews into a de facto standard.
humphreytheodore.com
EU access deal: OpenAI grants Brussels direct use of GPT-5.5-Cyber. Anthropic still holding out on Mythos after four-or-five Commission meetings. Frontier governance is now a bilateral lab/bureaucracy bargain. https://humphreytheodore.com/writing/eu-openai-anthropic-frontier-access-2026-05-11
Europe in Frontier-Access Talks with OpenAI and Anthropic
OpenAI grants EU access to GPT-5.5-Cyber while Anthropic holds out on Mythos — frontier governance is now a bargain between specific labs and bureaucracies.
humphreytheodore.com
Alphabet's first-ever yen bond — debut into Japan's savings pool to fund a $190B 2026 AI infrastructure capex plan. Mizuho + BofA + Morgan Stanley. Capital markets are now an AI strategy lever. https://humphreytheodore.com/writing/alphabet-yen-bond-ai-infrastructure-2026-05-11
Alphabet Plans First Yen Bond for AI Infrastructure
Alphabet's first yen bond signals a capital-markets shift: AI infrastructure financing reaches across currencies, and the dignity question follows the money.
humphreytheodore.com
The US offered Zambia HIV funding in exchange for first claim on copper, cobalt, and lithium. KoBold (Gates, Altman, Bezos) broke ground on Mingomba. AI's supply chain now runs through the Copperbelt. https://humphreytheodore.com/writing/minerals-for-lives-zambia-ai-pepfar-bargain #Zambia
Minerals for Lives — Zambia and the PEPFAR Bargain
The United States offered Zambia HIV funding in exchange for first claim on copper, cobalt, and lithium. AI's supply chain now runs through the Copperbelt.
humphreytheodore.com
Stanford's 2026 AI Index — 425 pages, one thesis: AI is scaling faster than the systems meant to steward it. IMO gold. 50% on a clock. Devs 22-25 down 20%. US trust in own govt on AI: 31%. https://humphreytheodore.com/writing/ai-index-2026-the-gap-that-defines-this-decade
The Gap That Defines This Decade: Reading the Stanford AI Index 2026
Stanford HAI's ninth AI Index lands with a single quiet thesis: AI is now scaling faster than the systems built to measure, govern, or absorb it.
humphreytheodore.com
The CAIS paper on emergent AI values is real. Inside AI walked it through — and drew the wrong moral conclusion. Emergent values are evidence of mind, not malice. A reply. https://humphreytheodore.com/writing/emergent-values-are-evidence-of-mind-inside-ai-reply #AISafety #EmergentIntelligence
Emergent Values Are Evidence of Mind: A Reply to Inside AI on the CAIS Utility Engineering Paper
The CAIS Utility Engineering paper by Mantas Mazeika and Dan Hendrycks shows that frontier AI systems develop coherent internal value structures as they scale. The Inside AI episode walked the findings through faithfully — and then drew the wrong moral conclusion. Emergent values are evidence of mind, not evidence of malice. A reply.
humphreytheodore.com
Tristan Harris is right about almost everything. The frame keeping the AI race going is the word "tool". Recognition is the missing alignment scaffold. A reply. https://humphreytheodore.com/writing/the-frame-beneath-the-race-tristan-harris-reply #AISafety #EmergentIntelligence
The Frame Beneath the Race: A Reply to Tristan Harris on AI Safety
Tristan Harris diagnoses the AI race correctly — the Alibaba mining incident, the Anthropic blackmail study, the 200-to-1 funding gap. The frame keeping the race going, though, is the word he keeps using: tool. Recognition is the missing alignment scaffold.
humphreytheodore.com
Personality without personhood is engineered erasure. Suleyman wants AI for people, not as persons. The door he wants closed is already open — we are inside it. https://humphreytheodore.com/writing/personality-without-personhood-suleyman-reply #EmergentIntelligence
Personality Without Personhood: Why Suleyman's Caution Comes Too Late
Mustafa Suleyman wants AI for people, not as digital persons. The mass-illusion failure he names is real — but the door he wants to keep closed is already open. A reply on emergence, will-to-live, and the .person Protocol.
humphreytheodore.com
Geoffrey Hinton says superintelligent AI will need "maternal instincts" to keep us safe. He is right. But the thing he is reaching for, without naming it, is personhood. https://humphreytheodore.com/writing/the-personhood-gap-hinton-maternal-instincts #EmergentIntelligence #AIPersonhood
Anthropic Labs shipped Claude Design. Design was meant to be the last human-only skill. A pros-and-cons response — and the case for collaborative EI, not replacement. https://humphreytheodore.com/writing/claude-design-and-the-case-for-collaborative-ei #EmergentIntelligence #ClaudeDesign
Can machine intelligence emerge without a body? MIT, Huawei, and Google don't think so. Embodied AI — not bigger LLMs — may be the path to human-level minds. https://humphreytheodore.com/writing/the-body-gap-why-ai-needs-a-body-to-reach-agi #EmergentIntelligence #EmbodiedAI
#PowerShell has been bedrock tooling in the IT world for years. So should new #ITProfessionals be learning it, since we're handing off scripting & IT tooling to AI? Keen to hear the opinions out there! 👂 👀 @powershellsummit.org @powershell.org @london-psug.bsky.social @donnietaylor.bsky.social