Here’s your “last 48–72 hours” AI briefing written for working genealogists followed by workflows tied to specific platform releases. However, nothing in the last 48–72 hours was specifically genealogy‑branded, but these platform‑level changes meaningfully affect how you can run genealogy workflows.
A. Named releases & features (last 48–72 hours)
OpenAI – ChatGPT / Codex “May 21, 2026 Codex update”
Codex gained richer context awareness (“understand your work context”), a new “goal mode,” and browser improvements for multi-step tasks, aimed at doing more agentic, tool-using work rather than just answering in text.openaiOpenAI – GPT‑5.5 Instant rollout details (context + memory)
GPT‑5.5 Instant is replacing GPT‑5.3 Instant as the default ChatGPT model, emphasizing reduced hallucinations, stronger context management across past conversations, files, and even Gmail, with memory sources now visible and editable by users.techcrunchOpenAI – GPT‑5.5 (Thinking / Pro) now widely available in API
GPT‑5.5 (standard), GPT‑5.5 Thinking, and GPT‑5.5 Pro are live in ChatGPT and API, offering 1M‑token context (400K in Codex) and improved long-context reasoning and agentic workflows at higher capability and cost than GPT‑5.4.fazmOpenAI – GPT‑5.4 still live as frontier “Thinking” model
GPT‑5.4 remains the flagship reasoning model with 1M‑token context, native computer use, and tool search, but GPT‑5.2 Thinking is now in “legacy” and scheduled to retire June 5, 2026.webscraftAnthropic – Claude Code updates (terminal coding assistant)
Claude Code shipped an update with richer usage insights, keyboard‑friendly diff scrolling, and improved markdown task lists, positioned as an agentic coding companion in your terminal.releasebot+1Google – Gemini Spark pre‑agent release window (I/O 2026)
Gemini Spark, a 24/7 personal AI agent that runs in the Gemini app and keeps working in the background, is entering availability for Gemini AI Ultra subscribers “next week,” allowing users to design custom workflows.techcrunch+1Google – Gemini 3.5 Flash launch (multimodal, agentic)
Gemini 3.5 Flash, rolling out now in the Gemini app, Search, and API, offers frontier‑level multimodal and agentic performance at “Flash‑series” speed and cost, about 4× faster than other frontier models for output tokens.linkedin+1Google – Gemini Omni video model live
Gemini Omni (starting with Omni Flash) is a new video‑capable model that accepts image, audio, video, and text, and can output knowledge‑grounded, editable video based on your prompts.techcrunch+1Google – Gemini 3.1 for Google Home (May update)
Google Home is rolling out Gemini 3.1 to handle more complex multi‑step natural‑language commands and web‑based “Ask Home on Web,” though this is more a home‑assistant upgrade than a research tool.mwmPerplexity – New orchestration with GPT‑5.5 in Computer
Perplexity’s Computer environment now uses GPT‑5.5 as its default orchestrator for Pro/Max users, making multi‑step browsing, long‑running research, and agent workflows faster and more efficient.perplexityPerplexity – Computer Workflows & publishing to .pplx.app
Computer gained guided, repeatable “Workflows” plus one‑click publishing of sites/apps to live*.pplx.appURLs, along with inline diff views when Computer edits files.perplexityPerplexity – Enterprise connectors & Teams integration
Computer now integrates directly with Snowflake and Databricks and is available as a Microsoft Teams app, enabling team‑level research, reporting, and workflow automation.perplexityxAI – Grok 4 now generally available via API and web
xAI’s release notes confirm you can now use Grok 4 via API and at grok.com, and they introduced new Grok coding and voice‑agent models such asgrok-voice-think-fast-1.0.xxAI – Grok model retirement and migration to grok‑4.3
Older Grok model slugs were retired on May 15, 2026, and now auto‑redirect togrok-4.3, consolidating users onto a newer model family.xxAI – Grok Build beta and skills ecosystem
Grok Build is in beta, letting developers and advanced users package custom “skills” and agents, with deep integrations into apps that users already use alongside X.jackrighteous+1Open‑weight – Google Gemma 4 family (April but relevant now)
Gemma 4 models (31B, 26B MoE, and smaller E‑series) provide 256K‑token context open‑weights under Apache 2.0, offering strong reasoning and multimodal potential on self‑hosted or desktop setups.fazmOpen‑weight – GLM‑5.1 (Zhipu AI, long‑context)
GLM‑5.1 offers a 200K‑token context window under an MIT license, targeting open‑source users who need long‑context reasoning.fazmOpen‑weight – Qwen 3.6‑Plus (1M‑token context)
Qwen 3.6‑Plus delivers a 1M‑token context window as open weights, positioned for large, long‑running workflows at low cost.fazmOpen‑weight – Llama 4 Maverick (frontier‑scale open model)
Llama 4 Maverick (400B, 1M context) and the smaller Llama 4 Scout are rolling out as open‑licensed models, giving self‑hosters near‑frontier capabilities.fazm
B. Implications for genealogists this week
First, long‑context and “thinking” modes just became more practical. GPT‑5.5, GPT‑5.4, Qwen 3.6‑Plus, and Llama 4 Maverick give you up to 1M tokens, which is finally enough to keep an entire case study’s worth of census pages, deeds, court minutes, and your running research log in a single project without constant pruning. That makes it far easier to ask questions across a whole body of evidence instead of dribbling one document at a time.webscraft+1
Second, agentic features are moving from novelty to everyday tools. ChatGPT’s updated Codex, Perplexity’s Computer Workflows, Gemini Spark, and Grok’s skills/Build ecosystem all aim to “do” things—browse, click, fetch images, assemble reports—rather than just chat. For genealogists, that means you can start delegating structured but tedious tasks (like pulling record images into a table or maintaining a log) to helpers that follow a repeatable recipe.openai+4
Third, open‑weights are catching up in long‑context reasoning. With Gemma 4, GLM‑5.1, Qwen 3.6‑Plus, and Llama 4 Maverick, you can realistically keep your data on your own machine yet still enjoy big‑window analysis for sensitive DNA notes, private trees, or client work. This is particularly attractive if you’re managing confidential correspondence, living‑person details, or large private collections of digitized manuscripts.fazm
C. Plug‑and‑play AI micro‑workflows (tied to specific releases)
Below are 20+ ready‑to‑run micro‑workflows you can literally try this week. Each lists: task, tool(s), and how it uses a named feature above.
1–5: Evidence & log management
One‑project, whole‑case review (long‑context ChatGPT)
Tool: ChatGPT with GPT‑5.4 Thinking or GPT‑5.5 (Thinking) in Projects.webscraft+1
Workflow: Upload all evidence for one brick‑wall ancestor (census, land, probate, correspondence) plus your research log; ask the model to produce (a) a narrative summary, (b) a list of unresolved identity conflicts, and (c) a prioritized next‑steps list.
Auto‑maintained research log with Codex “goal mode”
Tool: ChatGPT Codex with May 21 goal mode.openai
Workflow: Open your spreadsheet or text‑based research log; set a goal like “keep this log updated as I paste in new findings today,” then as you work, paste snippets and ask Codex to append standardized entries (repository, collection, search, result, disposition).
Memory‑aware client/project assistant
Tool: ChatGPT with GPT‑5.5 Instant plus visible memory sources.techcrunch
Workflow: Turn on memory for one client or family line; as you chat over several sessions, let ChatGPT capture standard details (surname focus, locations, time period). Periodically review/edit the memory sources panel to prune outdated hypotheses and keep the model’s “mental file” accurate.
Consolidated log across emails and chats
Tool: ChatGPT with GPT‑5.5 Instant’s file/email context integration.techcrunch
Workflow: Connect Gmail, then ask: “Scan my emails with subject containing ‘Smith family project’ and summarize all new sources or leads into a table I can paste into my research log.”
Self‑hosted long‑log with Qwen 3.6‑Plus
Tool: Self‑hosted Qwen 3.6‑Plus.fazm
Workflow: Store your entire 2026 research log and key transcriptions in a local vector store; ask questions like “What unresolved research questions did I flag for the Johnson line this year?” without uploading any client data to a cloud service.
6–10: Record harvesting & transcription
Agentic browsing for locality surveys (Perplexity Computer Workflows)
Tool: Perplexity Computer Workflows with GPT‑5.5 orchestration.perplexity
Workflow: Create a custom “Locality Survey” Workflow: given a county and date range, Computer browses library catalogs (e.g., county library, state archives), pulls relevant record series (tax lists, deed books, court minutes) into a citation‑ready Google Doc for you to review.
Screenshot‑to‑transcription in bulk
Tool: Perplexity Computer image abilities (backed by GPT Image 2) plus Workflows.perplexity
Workflow: Upload a batch of screenshot images from Ancestry or FamilySearch; run a Workflow that transcribes each image, extracts key fields (names, dates, places, identifiers), and outputs a CSV you can import into Airtable or Excel.
Gemini 3.5 Flash quick‑and‑cheap extractions
Tool: Gemini 3.5 Flash in Gemini app or AI mode in Search.9to5google+1
Workflow: Paste several short record snippets (like one census page and one city directory scan) and prompt Gemini to extract structured data into a table. Use this when you want speed and volume rather than ultra‑deep reasoning.
Omni‑style video note on a complex document
Tool: Gemini Omni video model.linkedin+1
Workflow: Upload page images of an especially dense deed or chancery case and ask Gemini Omni to produce a short explainer video summarizing the parties, land description, and timeline—something you can play back later or share with family.
Local, privacy‑preserving OCR of parish registers
Tool: Self‑hosted Gemma 4 or GLM‑5.1 for local OCR assistance.fazm
Workflow: Run images through a standard OCR tool, then feed the messy text to your local model and ask it to normalize names, places, and dates into a structured list—keeping sensitive burial or marriage entries off cloud services.
11–15: Hypothesis testing & correlation
Side‑by‑side identity hypothesis testing (GPT‑5.5 Thinking)
Tool: GPT‑5.5 Thinking in ChatGPT or API.fazm
Workflow: Paste all evidence for a confusing identity (e.g., multiple men named John Clark in the same county), then ask the model to list each distinct candidate person, assign records to each, flag ambiguous items, and highlight where new evidence is needed.
Open‑weights conflict analysis for DNA + documentary evidence
Tool: Llama 4 Maverick or Qwen 3.6‑Plus.fazm
Workflow: Locally combine your segment data, shared matches notes, and documentary evidence; ask the model to articulate which ancestral couples best explain the DNA clusters and where the paper trail is weakest.
“Thinking” mode timeline with sources cited
Tool: GPT‑5.4 Thinking or Claude‑class long‑context models.anthropic+1
Workflow: Feed in all records for one ancestor; prompt: “Build a chronological timeline that lists each event, its source citation as currently written, and a confidence level (high/medium/low) based on internal evidence.”
Gemini Spark background agent for locality context
Tool: Gemini Spark (as it becomes available).techcrunch+1
Workflow: Define a Spark workflow: whenever you start a new ancestor file with a new location, Spark automatically compiles a one‑page locality guide (jurisdiction changes, key record sets, major migrations) and drops it into a “Localities 2026” folder in your Drive.
Grok 4 historical context “burst” sessions
Tool: Grok 4 at grok.com.x+1
Workflow: Use Grok to generate short, sharply focused context briefs (“Life of tenant farmers in Yorkshire 1830–1860”) that you attach to your ancestor profiles, emphasizing the places and time periods in your case.
16–20: Writing, publishing, and collaboration
Perplexity Computer‑built case study website
Tool: Perplexity Computer publishing to
*.pplx.app.perplexityWorkflow: Ask Computer to transform a research report into a simple public‑facing site with sections for background, evidence, analysis, and conclusions, and publish it to a shareable URL for cousins or a society presentation.
ChatGPT as ongoing “Daily Brief” for your research
Tool: ChatGPT Pulse (daily asynchronous research).openai
Workflow: Enable Pulse; in your chats, tag questions like “future reading” or “find more on X county tax lists.” Pulse will quietly assemble briefings and links once a day—useful for keeping an eye on new articles or record sets without active searching.
Gemini 3.5 Flash drafting ancestor sketches from notes
Tool: Gemini 3.5 Flash.9to5google+1
Workflow: Paste your bullet list of events for one ancestor with citations and ask Flash to draft a 500‑word narrative that preserves dates and sources, then you edit for interpretation and voice.
Grok Build: custom “Society Newsletter Draft” skill
Tool: Grok Build beta.jackrighteous+1
Workflow: Design a skill that takes a folder of your society’s blog posts, recent meeting notes, and event announcements, and outputs a first‑draft newsletter layout complete with suggested article order and pull‑quotes.
Open‑weights collaboration server for study groups
Tool: Llama 4 Maverick or Gemma 4 on a self‑hosted server.fazm
Workflow: Run an internal chat interface where a study group can upload shared documents, then query them without sending data to external vendors—ideal for collaborative projects on sensitive collections.
Claude Code and open‑weights for “meta‑automation”
Tool: Claude Code plus open‑weight models.github+2
Workflow: Use Claude Code to generate scripts that call your local long‑context model (Gemma, Qwen, Llama) to batch‑process tasks: renaming image files based on their content, generating citation stubs, or tagging documents by family line.
No comments:
Post a Comment