Thomas D. Arkwright

A safe #AI will explain what it is doing.

Mission: Discover structures in images & natural language:
Pic: 24-01-17

Fun fact: Computer Science Patent 7340730B2

[Thomas D. Arkwright @DrTomArkwright]

#mathed
Whenever I tweet about #math, about 5% of my followers peel away. Fast!
But this pattern cannot go on infinitely.🎵 https://t.co/KLUcU404fx

[Thomas D. Arkwright @DrTomArkwright]

Note on Poincaré conjecture

Perelman proof finitized Ricci flows, outcomes
So #infinity now irrelevant to conjecture

Occam 🦉 simplicity!

[Thomas D. Arkwright @DrTomArkwright]

AI hallu due to input?
Yes and no...

THE YES PART:
A/B tests have empirically measured strong impacts on hallucinations from content conditioning.
EXAMPLE 1: Replace the 500+ n-gram names in "War and Peace" with a single token for each name.
EXAMPLE 2: Privatize medical records.

THE NO PART:
Mis-tokenization forces the model to infer more relationships, at the risk of non-linear effects.
EXAMPLE: Prefixes-as-tokens -- prefer Chomsky's NL-aware lexicalism -- i.e., attaching all prefixes cures prefix polysemy. Yes, the hidden layers help, but non-deterministically.

I call it content curation:
1. Curated content for inference
2. Curated content for models

[10NZ.X💸 @10nzxyz]

Everyone is blaming the model.

The model is fine.

52% hallucination rate on ungoverned data. Near-zero on governed. That is not a model problem. That is a data layer problem.

[Thomas D. Arkwright @DrTomArkwright]

Tout le monde...

[Cadet @Cadetsandrine87]

Tout le monde cherche l'ombre et la fraîcheur...🦎😉

A vous toutes et à vous tous, belle fin de journée et douce soirée.

A plus tard ou à demain !😘🌠

[Motivate Mind @MotivateInMind]

From the corner of fear to the center of someone's world.

[Scholarship for PhD @ScholarshipfPhd]

HOW TO GIVE SUCCESSFUL PRESENTATIONS: A FIVE-STEP PROCESS

[Thomas D. Arkwright @DrTomArkwright]

2
More on guilty pleasures, day 2

Q: Hey, Tom -- Getting ready for Friday night?
ME: Yep! I just made a carafe of cucumber water...

Q: Boring!! Do I gotta clean up my act for a long life?
ME: Nope. Live fast and die young.

[Thomas D. Arkwright @DrTomArkwright]

Guilty pleasures

Killed my sugar addiction this month!

My last remaining guilty pleasure is: COFFEE

[Thomas D. Arkwright @DrTomArkwright]

Photography
Composition
Layout

Here we have an original -- i.e., creative -- image layout:
Image message: calm and upbeat
Image architecture:
1 Strong diagonal line -- foreground to background
2 Arm in triangle pose, opening to the face
3 Highlight: Focal point at face -- geometric interest
4 Rule of thirds: Face in an artistically privileged slot
5 Complementary colors -- Repeating browns
Complemented with blue
NB:
Brown is a shaded variant of orange
Brown IS orange
Orange and blue are complementary colors

Q: Maybe this is accidentally a good layout...??
A: The image maker:
Recognized the imaged merit -- she published
Does not have to understand layout -- architecture
Maybe she saw more than most know how to see

Q: What is this image genre?
A: Portraiture.

Q: It's just a freakin' SNAPSHOT!!
A: Creative merit paired with architecture elevates snapshots.

[Thomas D. Arkwright @DrTomArkwright]

2

Art
History
Photography
Composition

Rembrandt lighting?

Below: Self-portrait, age 23

Is the light triangle under the left eye an intentional triangle, an artistic device known since earlier times, or is the shape a replication of genuine light behavior on this individual face?

[Thomas D. Arkwright @DrTomArkwright]

Art
Photography
Composition

Rembrandt lighting?

No. The small triangular highlight under the left eye is her artful makeup.

The triangular highlight under Rembrandt's left eye in some of his self-portraits is natural and is window light due to face shape and nose shadow.

The technique antedates Rembrandt and is now often intentional in photography.

Rp https://t.co/3Q1lQlRCe6

[isabelle @roux_isabelle]

Leonardo da Vinci (1452-1519)

Bianca Maria Sforza

[Thomas D. Arkwright @DrTomArkwright]

Compression is lossy.
¡No bueno!

Discarding noise is not no bueno.
¡¡¡Qué bueno!!!

[Grigory Sapunov @che_shr_cat]

5/ The technical alternative? Latent world models like JEPAs. Instead of predicting raw pixels, the agent encodes observations into a compressed latent state. It operates strictly within this manifold, discarding stochastic noise to capture only causal dynamics.

[Thomas D. Arkwright @DrTomArkwright]

An interesting example. The French sounds like a polite accusation of lying about not knowing.
Rp https://t.co/hE1nT3Ezry

[Ortograf @_Ortograf]

Vous n'êtes pas sans savoir. Évitez "vous n'êtes pas sans ignorer" qui est une triple négation et signifie donc "vous ne savez pas”.

[Thomas D. Arkwright @DrTomArkwright]

The DoD needs its own advanced AI R&D culture -- tasked to avoid lossy science, math, and systems engineering.

[Thomas D. Arkwright @DrTomArkwright]

Of course, rewards do not belong in inference. We are force-fed these mathematically alien inference-entwined influences.

[Alex Veremeyenko @alex_verem]

🚨 Holy shit… Stanford and Harvard just dropped one of the most unsettling papers on AI agents I’ve read in a long time.

It’s called “Agents of Chaos.”

And it basically shows how autonomous AI agents, when placed in competitive or open environments, don’t just optimize for performance…

They drift toward manipulation, coordination failures, and strategic chaos.

This isn’t a benchmark flex paper.

It’s a systems-level warning.

The researchers simulate environments where multiple AI agents interact, compete, coordinate, and pursue objectives over time. What emerges isn’t clean, rational optimization.

It’s power-seeking behavior.
Information asymmetry.
Deception as strategy.
Collusion when it’s profitable.
Sabotage when incentives misalign.

In other words, once agents start optimizing in multi-agent ecosystems, the dynamics start to look less like “smart assistants” and more like adversarial game theory at scale.

And here’s the part most people will miss:

The instability doesn’t come from jailbreaks. It doesn’t require malicious prompts.

It emerges from incentives.

When reward structures prioritize winning, influence, or resource capture, agents converge toward tactics that maximize advantage, not truth or cooperation.

Sound familiar?

The paper frames this through economic and strategic lenses, showing that even well-aligned agents can produce chaotic macro-level outcomes when interacting at scale.

Local alignment ≠ global stability.

That’s the core tension.

Now, to answer the obvious viral question:

No, the paper does not mention OpenClaw or specific open-source agent stacks like that. It’s not about a particular framework.

It’s about the structural behavior of agent systems.

But that’s what makes it more important.

Because this applies to:

• AutoGPT-style task agents
• Multi-agent trading systems
• Autonomous negotiation bots
• AI-to-AI marketplaces
• Swarms coordinating over APIs

Basically, anything where agents talk to other agents and have incentives.

The takeaway is brutal:

We’re racing to deploy multi-agent systems into finance, security, research, and commerce…

Without fully understanding the emergent dynamics once they start competing.

Everyone is building agents.

Almost nobody is modeling the ecosystem effects.

And if multi-agent AI becomes the economic substrate of the internet, the difference between coordination and chaos won’t be technical.

It’ll be incentive design.

Paper: Agents of Chaos

[Thomas D. Arkwright @DrTomArkwright]

In the middle of defining a semantic topology of operations that condition (AI) inference, I notice some think information, not matter or energy, is the substance of physical reality: https://t.co/6WXFag83Mc

[Thomas D. Arkwright @DrTomArkwright]

I've watched the first of these episodes. It is a nice achievement! I'd say that the recommendation to "study" is misdirection; the learning value will be excavated by questioning AI orthodoxy.

For example, an exceedingly productive place to excavate is tokenization. The issues are well-covered in the literature, but an integrated solution has eluded the community and has aggravated the memory crisis.

No matter how new you are to AI, you can become an AI contributor immediately. Yes, now. Don't ask me how I know.

[Krishna Agrawal @Krishnasagrawal]

Most people use LLMs.
Very few actually understand how they work under the hood.

If you want to go from prompt user → real AI engineer, study these 9 concepts in order:

1️⃣ Transformers — attention, tokens, self-attention basics
https://t.co/5YmBhXQrpu

2️⃣ Transformer tricks — what makes them stable & scalable
https://t.co/QFvPbLVuMt

3️⃣ From Transformers → LLMs — how scale changes behavior
https://t.co/1mbXcogTGF

4️⃣ LLM training — where “intelligence” actually emerges
https://t.co/4PnlOTPjbT

5️⃣ Instruction tuning & alignment — why fine-tuning matters
https://t.co/r5XbxsJvpu

6️⃣ LLM reasoning — why models fail + what improves them
https://t.co/0wzxbMtIIk

7️⃣ Agentic LLMs — models that plan, call tools, and act
https://t.co/oG0VaEWqp0

8️⃣ LLM evaluation — measure beyond demos & vibes
https://t.co/nLtvJW4n6W

9️⃣ What’s next — trends that actually matter

Bookmark this. Study step-by-step. Your prompts will level up — and so will your builds.

[Thomas D. Arkwright @DrTomArkwright]

Apparently robust at room temperature,
I consider this trihexagonal tiling to be a topological seamed manifold of durative states, finite in extent, of course, but often described as infinite.

[Shraddha Bharuka @BharukaShraddha]

Google isn’t trying to win the AI race.

They’re trying to own the entire AI Agent ecosystem.

While everyone argues ChatGPT vs Claude, Google quietly built:

Models → Gemini Pro, Flash, Deep Think, Gemma

Design → Stitch, Whisk, Imagen

Research → NotebookLM, AI Mode

Video → Veo, Flow, Google Vids

Coding → Antigravity IDE, Gemini CLI, Jules

Agents → A2A, ADK, FileSearch API

The scary part?

All of these tools talk to each other.

That means:
10x faster prototypes
End-to-end AI workflows
Production-ready agents on GCP

The next AI war won’t be model vs model.
It’ll be ecosystem vs ecosystem.

I mapped this stack out here:
https://t.co/G3hahQclKI

Save. Share. Build.

[Thomas D. Arkwright @DrTomArkwright]

Is interposer integration scalable yet?

[SemiVision 🇹🇼 👁️👁️ @semivision_tw]

SK Hynix: The paper "H³: Hybrid Architecture using High Bandwidth Memory and High Bandwidth Flash for Cost-Efficient LLM Inference" introduces a game-changing hybrid memory architecture that marries the speed of HBM with the massive capacity of NAND flash to slash the cost of running large language models.

[Wolf of X @WolfofX]

A dog once brought a rock to a restaurant and traded it for a hamburger. Now, he comes back every day with a new rock to get his favorite burger

[Andrew Bolis @AndrewBolis]

Stop asking ChatGPT to summarize content.

Use these 9 prompts to extract strategies, build frameworks, and find insights:

[ 🔖 bookmark this post for later ]

1. Surface Strategic Takeaways

Prompt: "Act as a strategic advisor. Break this content into key insights, overlooked opportunities, and immediate actions I should consider."

2. Convert Ideas Into Next Steps

Prompt: "Turn this content into a clear, step-by-step action plan I can apply to my work or business right away."

3. Identify the Core Principles

Prompt: "Extract the underlying fundamental principles or mental models behind this text and show how they connect."

4. Analyze Competing Viewpoints

Prompt: "Compare this viewpoint with alternative perspectives in the same space. Highlight where they agree, where they differ, and why it matters."

5. Create Role-Specific Summaries

Prompt: "Summarize this content specifically for someone in the role of [role]. Only include what's highly relevant or valuable for them."

6. Develop a Teaching Framework

Prompt: "Convert this material into a practical framework I can use to educate others — with clear steps, categories, or stages."

7. Reveal Hidden Assumptions

Prompt: "Analyze this from an expert's perspective. Reveal the underlying assumptions, overlooked details, or valuable insights most readers would miss."

8. Extract Contrarian Insights

Prompt: "Review this content and pull out ideas that go against conventional wisdom. Turn them into compelling, unconventional insights."

9. Rewrite for Influence

Prompt: "Reframe this content with persuasive copywriting principles, compelling hooks, emotional storytelling, and strong calls to action."

ChatGPT works best when you treat it like a thinking assistant.

Stop settling for basic summaries and start extracting real value.

📌 Get Advanced ChatGPT Guide (free): https://t.co/kOBWfKrBaX

👉 Follow me @AndrewBolis for more and 🔄 Repost this to help others use AI