Page 5 of 11
Listed inImpact on Product DevelopmentIntroductionon
Because the same input can produce different output, testing, cost, latency and failure handling all change shape. A feature also inherits its model's published retirement date, which makes migration scheduled work rather than a surprise.
Listed inCohereModels & Providerson
Reranking scores the query and document together instead of comparing two vectors, which is why it goes second and not first.
Listed inWeaviateVector Databaseson
A search service rather than a vector column or an embedded library. Hybrid search fuses keyword and vector ranking, which recovers the exact identifiers embeddings lose, and the index types include a flat one suited to many small per-tenant datasets.
Listed inConnect to a Local ServerModel Context Protocolon
Connecting over stdio is a process launch, not a connection. Absolute paths, the rule against printing to stdout, and where the logs are.
Listed inRegression TestingEvaluationon
Not "is this good" but "is this worse than last week". Build the set from failures you already had, and read the diff rather than the pass rate.
Listed inPre-trained ModelsHow LLMs Workon
There are two cutoff dates, not one, and the gap between them is where a model is most confidently wrong. Plus which symptom each stage explains.
Listed inOpenRouterModels & Providerson
One endpoint makes changing your mind cheap. The training opt-out governs the providers, not the router, and retention is not routed on at all.
Listed inRAG Use CasesRAGon
Four problem shapes retrieval fits and four it does not, why counting questions fail quietly, and how a million-token window moved the threshold.
Listed inContext IsolationContext Engineeringon
Running a sub-task in its own context window keeps its exploration out of the main conversation, which can return a short result from a long search. The same separation is a normative rule in the Model Context Protocol, and it costs the caller information the sub-task never reports.
Listed inLangSmithObservabilityon
A failing production trace becomes a saved test case, and that case runs against every later change. The documentation lists support for several frameworks and providers, so LangChain is not a prerequisite.
Listed inTop-KHow LLMs Workon
A fixed count applied to a distribution whose shape keeps changing. OpenAI never exposed it and Anthropic now returns a 400 for it.
Listed inClaude Messages APIAPIs & SDKson
The three required parameters include max_tokens, which is optional almost everywhere else. Content comes back as an array of typed blocks, so indexing the first one breaks the day you enable tools.
Listed inRAG with Dynamic FiltersRAGon
A forgotten metadata filter returns everyone's documents; a forgotten namespace returns an error. When the filter is a tenancy boundary, that asymmetry is the whole argument for a partition.
Listed inState & Historical ContextContext Engineeringon
Twenty turns into a booking, six settled facts are what the model needs and the transcript is an inefficient way to carry them. Trimming a turn per request can also cost more than sending everything, because it breaks the cache.
Listed inBias and FairnessSafety & Ethicson
You cannot inspect the training data, but your prompt, your examples and your retrieval are yours. Matched inputs that differ in one attribute test whether the system treats equivalent cases equivalently.
Listed inSystem PromptingPrompt Engineeringon
The Messages API takes the system prompt as its own parameter rather than a first message. Putting the date or a user's name in it stops the prefix caching, so you pay the write premium and never collect the read discount.
Listed inRecommendation SystemsEmbeddingson
Nearest neighbours of an item are a recommender you already have. It measures whether two things are about the same subject, which is not the same as predicting what a person will engage with.
Listed inTools & Function CallingAI Agentson
The model chooses a tool from its description alone, and every declaration is billed on every request whether or not it is called. Assume any tool with a side effect will be invoked twice.
Listed inImage GenerationMultimodal AIon
Rewriting the prompt gives you a different picture, not an edit of the one you had. Since no provider promises the same prompt returns the same image, store the result along with the prompt, the model and the interaction ID.
Listed inConstraining Inputs and OutputsSafety & Ethicson
Length caps control cost, enums keep invalid values out, and schemas prevent parse failures. None of them makes a value true, and a rule stated in a prompt is a request rather than a constraint.