@VoxelPerfect

AI Solution Architect | Leading Product & Technology for Enterprise AI Platforms | 2x Founder

 Athens, Greece
voxelperfect.com
Joined April 2010
  • Tweets1.3K
  • Following337
  • Followers215
50Media
I would evaluate an imported model through concrete controls: where data goes, who can inspect behavior, what can be switched, and how failures are contained. The governance argument from my July Kimi essay: voxelperfect.com/writing/the…

The Open-Weight Coalition: An AI Governance Answer to the Kimi Shock

Kimi K3 has triggered an organized industry response against broad restrictions on Chinese open models. The coalition is commercially interested—but its case for targeted security controls and...

voxelperfect.com
6
Hardware portability is a systems question. I would examine libraries, kernels, tooling, operations and switching cost before declaring a compute ecosystem replaceable. The argument from my July essay: voxelperfect.com/writing/cud…

CUDA Is Not Crumbling Yet: DeepSeek’s AI Strategy in Perspective

Liang Wenfeng is directionally right that AI and new compiler layers will weaken CUDA lock-in. But NVIDIA’s moat, the US compute advantage, and DeepSeek’s own model economics point to a more compli...

voxelperfect.com
3
What gets harder to copy after a year of using your AI product? My candidate: verified outcomes that improve evaluations, routing and workflow decisions. An operating loop, not just an inventory of data. voxelperfect.com/writing/the…
AI features are easy to copy. Learning from real use is harder. A teal ribbon curves around a worn coral block.

ALT AI features are easy to copy. Learning from real use is harder. A teal ribbon curves around a worn coral block.

3
NVIDIA can benefit from cheap, interchangeable models—if its compute platform stays hard to replace. Hugging Face helps with the first. Hardware portability complicates the second. The strategic tension behind the acquisition agreement: voxelperfect.com/writing/why…

Why NVIDIA Wants Models to Be Cheap

Joel Spolsky's old rule about commoditizing your complements explains NVIDIA's interest in Hugging Face—and exposes the harder question: can models become interchangeable without making the compute...

voxelperfect.com
7
A customer can stop depending on your product before cancelling it. I call that functional churn: the work leaves before the contract does. Which customer-health metric would notice? voxelperfect.com/writing/wha…
Customers can abandon your software before cancelling it. An empty teal office chair sits beside an open coral door and an aqua trail.

ALT Customers can abandon your software before cancelling it. An empty teal office chair sits beside an open coral door and an aqua trail.

6
I kept exploring a project despite better AI arriving later. My test: what will I understand even if I discard the implementation? A prototype can be temporary. The question it resolves need not be. voxelperfect.com/writing/i-d…
Waiting for better AI delays the learning you need now. An open notebook rests on discarded sketches against dark teal.

ALT Waiting for better AI delays the learning you need now. An open notebook rests on discarded sketches against dark teal.

6
A familiar entry point can be a distribution advantage for a new AI capability. My July essay used Chat, Work and Codex to ask when product consolidation makes discovery easier—and when it makes the experience confusing. voxelperfect.com/writing/ope…

OpenAI Is Turning Distribution Into the Product

By bringing Chat, Work, and Codex into one app and lowering the cost of trying them, OpenAI is turning its ChatGPT audience into a powerful distribution system for agentic work.

voxelperfect.com
17
A provenance signal is not an authorship verdict. The product question: does the interface preserve that distinction, or invite a stronger conclusion than the evidence supports? Revisiting my August essay: voxelperfect.com/writing/ant…
An AI watermark is not proof of who wrote the text. A teal rubber stamp casts a coral question-mark shadow on paper.

ALT An AI watermark is not proof of who wrote the text. A teal rubber stamp casts a coral question-mark shadow on paper.

21
A new capability can disappear inside an old evaluation framework. Before asking whether an AI product is good enough, ask: good enough for which job, measured against which assumptions? voxelperfect.com/writing/goo…
Public X exchange: Cheng Lou recalls Jeff Dean discussing Google’s internal pre-ChatGPT chatbot; Tibo Sottiaux describes his work on LMChat. Personal recollections, not a company-confirmed history.

ALT Public X exchange: Cheng Lou recalls Jeff Dean discussing Google’s internal pre-ChatGPT chatbot; Tibo Sottiaux describes his work on LMChat. Personal recollections, not a company-confirmed history.

1
11
Cheaper production does not decide who gets scarce energy, reliable compute or desirable land. Prices, quotas and access rules still allocate the bottlenecks. Abundance changes what money buys. voxelperfect.com/writing/abu…
AI can make work cheaper. It cannot abolish scarcity. A coral jug pours aqua water into a narrow teal funnel.

ALT AI can make work cheaper. It cannot abolish scarcity. A coral jug pours aqua water into a narrow teal funnel.

10
Jev was fastest on successful responses in our practical tests. It didn’t settle the choice. 77 intents, 17 questions per contract: batching helped every model. Cache hits changed costs; Jev had service failures. voxelperfect.com/writing/jev…
Specialized AI must beat the cheap model you already use. A coral pencil and teal drawing compass trace the same arc.

ALT Specialized AI must beat the cheap model you already use. A coral pencil and teal drawing compass trace the same arc.

18
Huang’s math remarks miss a distinction: a task can lose economic value while keeping educational value. Delegating work you understand differs from skipping the practice that builds understanding. voxelperfect.com/writing/jen…
AI can solve the problem. Students still need to learn the mathematics. An open exercise book contains a pencil-drawn geometric construction beside a coral pencil.

ALT AI can solve the problem. Students still need to learn the mathematics. An open exercise book contains a pencil-drawn geometric construction beside a coral pencil.

21
A coding agent can implement an unclear idea very quickly. The valuable skill is explaining what must stay true, what should change, and how you will check the result. voxelperfect.com/writing/cod…
Better coding agents make engineering judgment more valuable. A pencil casts the shadow of a measuring caliper.

ALT Better coding agents make engineering judgment more valuable. A pencil casts the shadow of a measuring caliper.

12
LLMs inherit culture through language. Fluent German does not guarantee reliable German reasoning. Aleph Alpha’s experiment makes that gap visible. Multilingual AI should be tested on local work, not translation quality alone. voxelperfect.com/writing/llm…
LLMs inherit culture through language. An open book fans into cream, teal, and coral pages with faint landscapes, architecture, and botanical imagery.

ALT LLMs inherit culture through language. An open book fans into cream, teal, and coral pages with faint landscapes, architecture, and botanical imagery.

12
Claude Marketplace borrows a familiar Azure playbook: let partner purchases use an existing commitment. That can make switching models a budget question as well as an engineering question. Anthropic's purchasing path remains a limited preview. voxelperfect.com/writing/cla…
AI lock-in can start in procurement, not code. A teal wallet and coral shopping bag linked by an aqua ribbon.

ALT AI lock-in can start in procurement, not code. A teal wallet and coral shopping bag linked by an aqua ribbon.

1
1
26
LLMs may not have human emotions because they do not have human bodies or human lives. But “not like us” is not the same as nothing. Anthropic found emotion-like internal states in Claude that causally change preferences and behavior. voxelperfect.com/writing/llm…
Three distinct claims: observed emotional language, causal evidence for functional emotion, and unknown subjective feeling. Evidence for one layer does not settle the next.

ALT Three distinct claims: observed emotional language, causal evidence for functional emotion, and unknown subjective feeling. Evidence for one layer does not settle the next.

18
Gen Z uses AI while trusting it less. That is not technophobia. It may be the first post-naive-tech generation: raised inside platforms after surveillance, attention engineering and misinformation were already visible. Trust is now part of the product. voxelperfect.com/writing/gen…
Gallup survey comparison for U.S. Gen Z, 2025 to 2026: weekly AI use stays at 51 percent, excitement falls from 36 to 22 percent, and anger rises from 22 to 31 percent.

ALT Gallup survey comparison for U.S. Gen Z, 2025 to 2026: weekly AI use stays at 51 percent, excitement falls from 36 to 22 percent, and anger rises from 22 to 31 percent.

9
OpenAI's Navier–Stokes claim still needs scrutiny. Tao is right that answers alone aren't understanding. My response: fund explanation and experimentation, then expand the research agenda. AI should raise our scientific ambition. voxelperfect.com/writing/ai-…
One promising question branches into a family of hypotheses. The headline reads Keep the rigor. Expand the search. Each hypothesis still needs evidence. A conceptual illustration.

ALT One promising question branches into a family of hypotheses. The headline reads Keep the rigor. Expand the search. Each hypothesis still needs evidence. A conceptual illustration.

11
Figure pays people to record tasks for robot training. IKEA reskilled 8,500 call-centre workers for design and complex service as its chatbot handled routine enquiries. AI automates tasks, not collars. Leaders choose what happens next. voxelperfect.com/writing/the…
Figure source photo of people recording a physical task. Index reports 44,000 weekly creators, 16 million task videos and 15 million dollars paid to creators.

ALT Figure source photo of people recording a physical task. Index reports 44,000 weekly creators, 16 million task videos and 15 million dollars paid to creators.

16
Every agent should not become another software project with its own runtime, evals, permissions and deployment. Let the platform own the repeated engineering. Let the agent live as a governed capability inside it. voxelperfect.com/writing/don…
An agent is shown inside Gaia's governed runtime, surrounded by tools, context, workflows, evaluation, and governance, with no separate agent deployment.

ALT An agent is shown inside Gaia's governed runtime, surrounded by tools, context, workflows, evaluation, and governance, with no separate agent deployment.

11
Load more