Here’s what actually changed in the last 48–72 hours plus a few still‑active rollouts from the past few days that matter in practice.
First, some models are quietly going away while newer, more capable “thinking” models with huge context windows are becoming the norm. Practically, that means your best bet for complex genealogy work is to deliberately choose the most recent “Thinking” or “Opus 4.8/Sonnet 4.6” style models when you need careful reasoning over many documents, and stop depending on GPT‑4.5 or GPT‑5.1‑era custom prompts.
Second, both OpenAI and Anthropic are leaning into sustained, long‑running workflows: 1M‑token contexts, managed agents, better prompt caching, and improved handling of long pastes/attachments are all signals that they expect real users to dump large working sets into a model and keep coming back over days. For genealogy, that’s an invitation to treat a model as a persistent analysis partner for a single research problem—say, a Muskogee probate cluster or a multi‑generational Cherokee family—rather than a series of one‑off chats.
Finally, AI news sources continue to show that general‑purpose agents and specialized security models (like Claude Mythos) are being tested in demanding domains, which indirectly reassures us that agentic patterns are maturing. In genealogy, we’re right on the cusp of similar “agentic” workflows: assistants that can read whole project folders, traverse trees and catalogs, and maintain a rich, long‑term picture of your research question, not just a single record at a time.[
A. Named releases & features (last ~72 hours)
OpenAI – ChatGPT: GPT‑4.5 retirement date announced
OpenAI has set June 27, 2026 as the retirement date for GPT‑4.5 inside ChatGPT, nudging users toward newer GPT‑5.x models for everyday work.[instagram][youtube]OpenAI – ChatGPT: GPT‑5.5 Instant update (May 28, now fully live)
GPT‑5.5 Instant, the default model in ChatGPT, was updated to give tighter, more readable answers and to remove the old “canvas” mode in favor of richer writing and code blocks directly in chat.[youtube]OpenAI – ChatGPT: Deep reasoning with GPT‑5.4 Thinking (earlier, but key this week)
GPT‑5.4 Thinking offers a longer, 256k‑token context and explicit “thinking” mode, making it better at long, multi‑document reasoning and deep web research.[youtube]OpenAI – ChatGPT: File Library expansion to Free/Go (rollout still in progress)
File Library (persistent storage for uploads) now includes Free and Go users, letting you keep and re‑use PDFs, spreadsheets, and images across chats up to your storage limit.[youtube]OpenAI – ChatGPT: Large pastes become attachments (Plus/Pro/Business)
Pasting more than about 5,000 characters automatically converts the text into an attachment instead of flooding the message box, preserving context window for analysis.[youtube]Anthropic – Claude: New Claude Opus 4.8 general model (May 28)
Claude Opus 4.8 is Anthropic’s newest top‑end model with a default 1M‑token context window, high‑resolution vision, adaptive “extended thinking,” and support for tools like web search and computer use.[youtube]Anthropic – Claude: Advisor tool upgrade (June 2)
The “advisor tool” now supports a max_tokens setting so you can cap how verbose the advisor model is when guiding another, faster model.[youtube]Anthropic – Claude: Refusal billing fix (June 2)
Requests that end in a hard refusal without any generated text are no longer billable on the Claude API.[youtube]Anthropic – Claude: Outage + recovery on June 2 (Opus 4.6 focus)
Claude had a partial outage on June 2 affecting some Claude.ai and API usage (notably Opus 4.6), with service restored after Anthropic deployed a fix.[techradar]Anthropic – Claude: Managed Agents, multi‑agent, and workflows on AWS (late May, still expanding)
Claude’s fully managed agents, multi‑agent orchestration, self‑hosted sandboxes, and webhooks are now available on Anthropic’s AWS‑hosted platform, making “agentic” workflows easier to deploy.[youtube]Anthropic – Claude: Opus 4.8 fast mode (research preview)
Fast mode for Opus 4.8 allows significantly faster generations at premium pricing, geared toward long‑running or interactive tasks.[youtube]OpenAI – ChatGPT: Job search & resume tools (June 1)
ChatGPT now offers US‑only live job listings and an integrated resume‑formatting experience, with global English‑language resume formatting.[youtube]OpenAI – ChatGPT: Memory sources & more personalized responses (May 5, still rolling out)
Memory “sources” now show which past chats, memories, or files informed a response, and GPT‑5.5 Instant can better use those for continuity.[youtube]OpenAI – ChatGPT: ChatGPT for Excel and Google Sheets add‑ins (May 5)
Official add‑ins embed ChatGPT in Excel and Sheets sidebars, letting the model read and write directly in your spreadsheets.[youtube]OpenAI – ChatGPT: Expanded “Thinking” context (Feb 20, but crucial for long projects)
Thinking mode now supports a total of 256k tokens (128k in / 128k out), useful for very large document sets.[youtube]Anthropic – Claude: 1M‑token context now GA on Sonnet 4.6 & Opus 4.6 and 4.8
A 1M‑token context window is generally available for key models, enabling truly massive multi‑document sessions (with long‑context pricing).[youtube]Anthropic – Claude: Structured outputs and text editor tools
Structured JSON outputs and an updated text editor tool allow Claude to produce machine‑readable summaries and edit long text files programmatically.[youtube]OpenAI – ChatGPT: Retirement of older models (GPT‑5.1 and earlier)
GPT‑5.1 models and older GPT‑4‑era variants have been retired in ChatGPT, concentrating improvements in GPT‑5.3, 5.4, and 5.5 series.[youtube]Meta / Google / Open‑weight: Gemma 4 open‑weight models for reasoning & agents (late May)
Google’s Gemma 4 open models, designed for advanced reasoning and agentic workflows, are now available for self‑hosting and fine‑tuning.[crescendo]
B. Plug‑and‑play AI micro‑workflows for genealogists (tied to current features)
Below are 20+ concrete micro‑workflows you can drop into your practice today, each taking advantage of something in the release list above.
1–5: Long‑context evidence reviews (Claude Opus 4.8 & Sonnet 4.6, GPT‑5.4 Thinking)
“One‑surname, one‑problem” evidence digest
Tool(s): Claude Opus 4.8 or Sonnet 4.6 with 1M‑token context; GPT‑5.4 Thinking with 256k tokens.[youtube]
Workflow: Upload a compiled PDF of everything on one troublesome ancestor (probate packets, land descriptions, census images, compiled notes); ask for a structured table: source, date, locality, principal parties, informant, and assessment of reliability.[youtube]
Locality study from mixed files
Tool(s): Claude Opus 4.8 / Sonnet 4.6 extended context.[youtube]
Workflow: Combine county histories, territorial statutes, and an inventory of record types into one massive document; prompt the model to identify record gaps and time periods where your Oklahoma county record coverage is weak.[youtube]
Conflicting‑evidence matrix for a single identity
Timeline harmonization across multiple families
Bulk translation and pattern spotting in foreign records
6–9: Smarter “document dump” via attachments & File Library (ChatGPT + Claude)
Paste‑to‑attachment cleanup for messy research logs
Tool(s): ChatGPT with large pastes auto‑converted to attachments.[youtube]
Workflow: Paste an overlong research log; let ChatGPT convert it to an attachment; then ask for: “Standardize and deduplicate this log; output a clean table I can paste into Excel (columns: date, repository/site, search terms, result, next action).”[youtube]
File Library as a genealogy project shelf
Tool(s): ChatGPT File Library (Free/Go/Plus/Pro, rolling out).[youtube]
Workflow: Upload a small bundle per project (research plan, log, working narrative, key deeds/probates) and label files with project codes; in a new session, say “Open the files for project MORGAN‑MUSKOGEE and summarize what we concluded last time plus three next actions.”[youtube]
Persistent “problem file” for a brick wall
Tool(s): ChatGPT File Library + GPT‑5.5 Instant.[youtube]
Workflow: Keep a single evolving “BrickWall_John_Doe” doc in Library; after each research day, upload the updated version and ask the model to: “Append a short ‘current hypothesis’ and ‘to‑test next’ section, dated today, based on this file.”[youtube]
Claude structured‑output index of an uploaded PDF book
10–13: Spreadsheet‑centric workflows (ChatGPT for Excel/Sheets)
Automated FAN‑club table building in Excel
Probate inventory categorization and valuation
“Where have I already looked?” log normalizer
Citation quality scanner in Sheets
14–16: Agentic and “managed” workflows (Claude Managed Agents & Opus 4.8)
Autonomous locality‑catalog digger
Tool(s): Claude Managed Agents with web search and code execution (on Claude Platform / AWS).[familylocket][youtube]
Workflow: Configure an agent with instructions to search a specific locality (e.g., “Muskogee County, Oklahoma, pre‑statehood court records”) in FamilySearch Catalog, state archives, and major libraries, then output a Google‑Doc‑style list of relevant collections with call numbers and access notes.[familylocket][youtube]
Mass‑record‑set summarizer for one family line
Tool(s): Claude Managed Agent + 1M‑token Opus 4.8.[youtube]
Workflow: Point the agent at a folder of transcribed deeds, tax rolls, and probate extracts for one surname; let it run a multi‑step plan to extract per‑person timelines and a consolidated land‑parcel map keyed to legal descriptions.[youtube]
“Advisor + executor” pattern for research planning
Tool(s): Claude advisor tool with max_tokens; Sonnet/Opus as executor.[youtube]
Workflow: Use the advisor to draft a concise, theory‑rich research plan for a problem (limited to, say, 400 tokens), then have the faster model convert that into a bulleted research task list you can paste into your log or project plan.[youtube]
17–20: Model retirement & continuity (“migrate your prompts”)
Update your old GPT‑4.5 prompt library to GPT‑5.x
Tool(s): GPT‑5.5 Instant and GPT‑5.4 Thinking; note GPT‑4.5 retirement on June 27.[instagram][youtube]
Workflow: Paste an old curated genealogy prompt into GPT‑5.5 Instant and ask: “Rewrite this for GPT‑5.4 Thinking with explicit steps, better safety language, and references to attachments instead of inline pastes.” Repeat for your favorite “record checklist” and “conflict table” prompts.[instagram][youtube]
Back‑test a key brick‑wall argument on the new models
Consolidate multi‑chat histories using Memory Sources
Tool(s): ChatGPT memory sources + GPT‑5.5 Instant.[youtube]
Workflow: If you have many prior chats about the same research problem, ask: “Using memory sources, summarize everything we’ve discussed about my [SURNAME] line in Muskogee, listing outstanding questions and untested hypotheses.” Then manually verify the summary against your own notes.[youtube]
Structured export for integration with Zotero or RootsMagic
Tool(s): Claude structured outputs; Gemma 4 self‑hosted if you prefer local.[crescendo][youtube]
Workflow: After the model analyzes a set of notes, ask it to output a JSON or CSV formatted list of events (date, place, person(s), source ID, note) that you can script‑import into your genealogy database or Zotero collections.[crescendo][youtube]
Job‑search feature repurposed as “repository‑search” practice
Tool(s): ChatGPT job search module (US only) + resume formatter.[youtube]
Workflow: As a practice exercise (or if you do paid research): use the job search to locate remote genealogy‑related roles or contract positions, then feed the postings into the resume formatter and ask: “What genealogical skills and record‑types are most in demand in these postings, and what gaps do I have?” This can guide where to invest learning time that helps both your own research and any client work.[youtube]
No comments:
Post a Comment