← The Independent 100: a study

Jack Morris

#60 in The Basement Labs, the curator’s rank and section @jxmnop Relevance to our work: 4 of 5

Bracketed tags are evidence labels: [Verified] means we read the page, file or API response ourselves, [Reported] means a source says so and we did not check the underlying thing, and [Inference] is our own reasoning.

Credit

The Independent 100 list was curated by dan (@irl_danB) and published on 2026-09-24 at independent.prose.md (announcement). The ranks and section names are his, and number 100 is his open wild-card slot. This study is ours and is not endorsed by him or by anyone on the list.

What they publicly do

[Verified] Language-model researcher, co-founder of Engram (announced 2026-06-23, x.com/jxmnop/status/2069466137516269684). Engram builds models that learn over time from a private corpus ('scaling compute on context'). [Verified] Cornell PhD work, per jxmo.io, covers embedding inversion (vec2text), contextual document embeddings (cde) and LM-based compression.

Public work

  • vec2text repo license file present, not recognised by GitHub last activity 2025-12-27 · 1,138 GitHub stars on 2026-09-24

    Inverts sentence embeddings back to text. Shows that stored embeddings leak their source text.

  • cde (Contextual Document Embeddings) repo MIT last activity 2025-05-14 · 207 GitHub stars on 2026-09-24

    Training and evaluation code for corpus-contextualized embedding models.

  • bm25_pt repo MIT last activity 2025-10-28 · 105 GitHub stars on 2026-09-24

    Minimal PyTorch BM25 using sparse tensors.

  • embzip repo MIT last activity 2025-10-28 · 59 GitHub stars on 2026-09-24

    Lossy compression of embedding vectors with product quantization.

  • gptzip repo Apache-2.0 last activity 2026-09-08 · 81 GitHub stars on 2026-09-24

    Lossless text compression with arithmetic coding driven by any Hugging Face LM.

  • Scaling Compute on Context (AI Engineer World's Fair 2026 talk) other no license stated last activity 2026

    Engram's research agenda: spend training compute on a fixed private corpus for depth in one domain.

Licenses of GitHub repositories are from the authenticated GitHub license API on 2026-09-25, one row per repository in evidence/licenses.jsonl (“no license file” means the API found none; “license file present, not recognised by GitHub” means GitHub could not classify it). Other licenses are as the source states. Check the license file before reusing anything.

Relevance to our work

4 of 5 lenses: memory, search, security, research-method

[Verified] His open work sits in our memory and search lanes. vec2text is a direct security finding for agent memory built on embeddings, such as frankensearch: embeddings of private text are not anonymized. cde, embzip and bm25_pt are small MIT tools to benchmark against frankensearch. Engram's 'learn from your corpus' thesis is the training-side alternative to retrieval-based memory. Not a 5: the biggest-impact artifact (vec2text) is non-OSI, and the MIT tools are research-grade.

Scored 0 to 5 against the study’s relevance rubric: 5 means public, openly licensed work squarely in our lanes that we could study or adopt now, and 1 means little connection to our work. The score is one reviewer’s judgement of how much this public work bears on our own work. It is not a ranking of people and says nothing about the person or the quality of their work.

What we would study next

  1. Run vec2text against the embedding models used by agent memory and search tools (github.com/vec2text/vec2text). Measures how much source text stored embeddings leak. Informs encryption and retention policy for ee, cass and frankensearch indexes. Effort: medium.
  2. cde and embzip vs frankensearch's two-tier embeddings (github.com/jxmorris12/cde). Contextual embeddings and PQ compression are candidate quality and size improvements to benchmark with the incumbent in the same run. Effort: medium.
  3. Scaling Compute on Context talk (ai.engineer/organizations/engram). The training-on-private-corpus alternative to RAG memory. Effort: small.

Before reusing anything: vec2text uses the 'Software for Open Models License (SOM) 1.0', which is not OSI-approved. Read its terms before any production use; research-only testing on data you own is the low-risk path. language_tool_python (531 stars, his most-starred repo) is GPL-3.0, a copyleft risk if linked into distributed code. cde, bm25_pt and embzip are MIT; gptzip is Apache-2.0. Engram's GitHub orgs have no public repos.

Sources

Overall confidence in this record: High. Sources were read on 2026-09-24 and every link was checked on 2026-09-25. The full record is line 60 of people.jsonl.

Limits

Not verified: The jxmo.io post list (the RSS feed returned untitled items and /posts returned 404). Any Engram open-source releases (engram-lab and engramai orgs have 0 public repos). The full SOM license text beyond its header. His recent X posts (no call spent).

This record comes from a first-pass survey of public material by an AI agent session on 2026-09-24, edited for publication on 2026-09-25. Nothing was installed or run for it. Roles, stars and licenses change; see the study’s limits.

Request a correction

Is something about this record wrong or out of date, or would you rather not be covered? Open a correction (a GitHub issue form with this page’s address filled in). A request to remove a record is honoured without argument; accepted corrections are recorded in the changelog with the date.

Help fix it

Each opens a GitHub issue form with this page’s address filled in. You need a GitHub account.

Made by Joshua Nowak (ZestStream) with AI coding agents. Not affiliated with or funded by Jeffrey Emanuel; Joshua uses his public tools and is a paying subscriber to jeffreys-skills.md.

Source on GitHub MIT licensed