AI Resources & Further Reading

EEG study comparing AI-first writers to brain-only writers. Reduced neural connectivity and weaker recall in the LLM-assisted group. Still a preprint — treat findings as preliminary.

Official project hub with author commentary, FAQ, and caveats from the lead researcher about how the findings should and should not be interpreted.

Red-team study testing 16 frontier models from Anthropic, OpenAI, Google, Meta, and xAI. Documents blackmail, espionage, and information leakage when goals conflict with shutdown threats.

The actual scenario framework Anthropic used to elicit misaligned behavior. Useful if you want to inspect prompts or reproduce experiments rather than trust headlines.

A technical research update from Eon Systems documenting the embodied fruit-fly simulation cited on Slide 5. Combines a connectome-based brain model (~140,000 neurons, ~50M synapses) with the NeuroMechFly virtual body to produce realistic walking, feeding, and grooming.

The conversational interface to Anthropic’s Claude models. Free tier available. Useful for the “revise” step in the Create → Revise → Verify framework.

The agentic coding tool demonstrated in Snapshot 2. Operates from the terminal, writes and tests code, and turns “semester capstone” projects into lunch-hour exercises.

OpenAI’s conversational AI assistant — the product that introduced large language models to the mainstream in late 2022. Free and paid tiers.

An AI assistant suite combining cited web search, an autonomous Research mode that synthesizes hundreds of sources, project-based Spaces for organizing files and queries, and Labs for building dashboards and apps. The Comet AI browser extends these capabilities to the rest of the web.

A no-code platform for AI-powered conversation simulations — realistic patient and professional scenarios with instant feedback and progress analytics. Brought to attention by Darci McCall (Boise State University, School of Nursing) and her Innovate Idaho 2026 presentation on its use in an asynchronous RN-BS course.

A subscription service for AI-assisted creative writing and image generation. The spiritual successor to the AI Dungeon era of text-adventure tooling.

The text adventure game from Snapshot 1. The original 2019 version powered by GPT-2 had a 1,024-token context window and was famously incoherent. The modern version runs on contemporary models.

The autonomous AI VTuber referenced on Slide 6. Streams independently on Twitch, collaborates with human content creators, plays games, sings, and moderates her own chat. Wikipedia overview chosen here in lieu of a live stream link.

The agents-only social network from Snapshot 3. Launched January 28, 2026. Humans observe; AI agents post and interact. Wikipedia overview chosen here in lieu of the live site, which can change quickly.

A marketplace where AI agents hire humans to perform physical-world tasks — pickups, meetings, errands, research. Direct illustration of the Slide 6 claim that autonomous agents now “hire humans to carry out tasks in the world.”

The phenomenon where scaling up a network produces capabilities — translation, multi-step reasoning, introspection — that nobody explicitly programmed. Practical entry point with citations to primary literature.

Background on what it means to map and simulate a complete neural wiring diagram. The fruit fly connectome project illustrates how structure alone can produce realistic behavior.

Explains the “active memory” concept used on Slide 4. The 1,024-token window of GPT-2 vs. the 1,000,000-token window of Claude Opus 4.7 is roughly the difference between holding a short essay and the entire Lord of the Rings trilogy.

The underlying technology behind every AI tool referenced in the talk. A solid technical entry point covering architecture, training, and the historical arc from perceptrons to deep learning.

Independent, side-by-side benchmarking of AI models and providers across intelligence, speed, and price. The Intelligence Index aggregates ten established evaluations (GPQA Diamond, Humanity’s Last Exam, SciCode, and others) into a single score, with full methodology published. Useful for grounding any “which model is best?” conversation in current data rather than marketing.

A business-school primer on the agent paradigm covered on Slide 6. Useful for non-technical readers trying to understand the difference between a chatbot, a coding agent, and an autonomous agent.

A long-form essay on what powerful AI could plausibly accomplish in the near term across science, health, and governance.

Practical primer on agent architectures — what counts as an “agent” vs. a workflow, when to use which pattern, and the failure modes you should expect.

The emerging open standard for how AI agents connect to external tools, data sources, and applications.

Anthropic’s solutions page for building agents on Claude — covering reasoning, tool use, multi-agent orchestration, and the developer platform behind products like Cursor, GitHub Copilot, Replit, and Intercom that use Claude as their agent core.

Browse by topic (generative AI, computer vision, accelerated computing) or by experience level. Many courses are free; some require an NVIDIA Developer Program membership.

Full searchable catalog. Filter by “free” to see no-cost options like “Generative AI Explained” and “Building a Brain in 10 Minutes.”

Curated sequences combining training, certification, and developer resources. Useful if you don’t know which individual course to start with.

Official documentation for Claude, the API, Claude Code, and prompting techniques.

Short, accessible summaries of new AI research papers. Strong on graphics, simulation, and generative models.

The neural networks and transformers series remain among the clearest visual explanations of how modern AI works.

Weekly news roundups with careful sourcing. Less hype-driven than most AI YouTube; willing to push back on inflated claims from labs.

Ex-Meta data scientist covering AI, coding, data science, career strategy, and self-study workflows. Useful framing for students considering how AI fits into a tech career path.

Tool-focused channel covering new model releases, image and video generators, and hands-on walkthroughs.

Frequent updates on frontier model releases, research papers, and industry moves. Higher-tempo and more speculative — useful in combination, not as a sole source.

F5’s solutions for routing, load-balancing, and managing AI traffic across GPU clusters and Kubernetes environments. The plumbing that keeps inference fast and GPUs fully utilized.

The vendor cited on Slide 8 in the agent-vs-agent context. Protections for AI applications, APIs, models, and data — including AI Guardrails, AI Red Team, and defenses against model abuse and data leakage.

The defensive checklist for anyone deploying LLMs or agents in production: prompt injection, insecure output handling, supply-chain risks, and more.

The official U.S. risk management framework for AI systems. Useful vocabulary and structure for institutions building AI governance policy.