2023¶

August 22, 2023
in Knowledge Strategy, AI Engineering, AI Governance
3 min read

Retrieval is a knowledge-governance problem

International Business Machines (IBM) Research has published a clear explanation of retrieval-augmented generation, an approach that gives a large language model access to external sources at the time of a request. Instead of relying only on patterns encoded during training, the system retrieves relevant material and uses it to generate a more current, domain-specific, and potentially verifiable answer.

Retrieval-augmented generation (RAG) is a promising architecture. It is not a substitute for governing the knowledge being retrieved.

August 15, 2023
in Defense Technology, Systems Engineering, Innovation Management
3 min read

Lunar infrastructure starts with interoperability

The Defense Advanced Research Projects Agency (DARPA) has announced the 10-Year Lunar Architecture (LunA-10) capability study. The effort will examine how a future lunar economy might move from isolated, self-sufficient projects toward shareable and scalable infrastructure for power, communications, logistics, and other services.

The timing of that architectural work matters. Interoperability is easiest to praise before systems exist and hardest to achieve after every provider has optimized a private solution.

August 9, 2023
in Cybersecurity, Defense Technology, Innovation Management
3 min read

AI cyber defense needs a transition path

The Defense Advanced Research Projects Agency (DARPA) has launched the Artificial Intelligence Cyber Challenge (AIxCC), a two-year competition aimed at using artificial intelligence to find and fix vulnerabilities in widely used software. Anthropic, Google, Microsoft, OpenAI, and the Open Source Security Foundation are participating in the effort, with competitions planned around DEF CON.

The challenge is ambitious for good reason. Software underpins critical infrastructure, and human defenders cannot manually inspect every dependency at the speed new vulnerabilities appear. The difficult work will begin when a winning technique meets a real maintainer's backlog.

July 31, 2023
in Defense Technology, AI Assurance, Innovation Management
3 min read

Trustworthy AI needs a research agenda, not a slogan

Researchers from academia, industry, and government are gathering this week for the Defense Advanced Research Projects Agency's Artificial Intelligence (AI) Forward workshop. The agenda centers on a question that is easy to state and difficult to engineer: how can AI systems operate reliably, interact appropriately with people, and support national-security needs under demanding conditions?

Calling a system trustworthy does not make it so. Trustworthiness has to be decomposed into research questions, engineering evidence, and operational learning.

July 25, 2023
in AI Security, National Security Technology, Risk Management
3 min read

Frontier-model security is shared infrastructure

Anthropic has published an initiative focused on the security of advanced artificial intelligence models. The concern is straightforward: as models become more capable and expensive to produce, their weights, training systems, research, and deployment infrastructure become valuable targets for theft or misuse.

The security problem does not sit in one server room. It crosses the organization and its supply chain.

July 18, 2023
in AI Strategy, Platform Engineering, Enterprise Architecture
3 min read

Open models turn selection into engineering

Meta and Microsoft have announced commercial access to Llama 2, with support across Azure and Windows. The release expands the range of models organizations can host, adapt, and integrate under their own architectural choices rather than consume only through a closed service.

That optionality is valuable. It also moves more of the responsibility from procurement into engineering.

July 11, 2023
in Human-Centered AI, Knowledge Strategy, Product Management
3 min read

Long context changes the knowledge-work interface

Anthropic has released Claude 2 with a context window that can accept roughly 100,000 tokens—enough for hundreds of pages of material in one prompt. The immediate attraction is obvious: a user can bring a long report, technical documentation, or even a book into a conversation without dividing it into tiny fragments.

More context changes what a language model can see. It does not guarantee that the model will attend to the right thing.

July 6, 2023
in Platform Engineering, AI Governance, Product Management
3 min read

Stable model access is part of the control plane

OpenAI has made the Generative Pre-trained Transformer 4 (GPT-4) application programming interface generally available to existing paying developers and announced a retirement path for several older completion and embedding models. The two developments belong together. Production access is not only about opening capacity; it is about managing change.

When an external model becomes part of an application, version stability and migration become product concerns.