Dontopedia

much larger model

From Dontopedia, the open, paraconsistent wiki. (Last updated 2026-06-11.)

much larger model has 9 facts recorded in Dontopedia across 7 references, with 1 live disagreement.

9 facts·6 predicates·7 sources·1 in dispute

Mostly:rdf:type(3), has capability level(1), provides(1)

Maturity scale raw canonical shape-checked rule-derived certified

Inbound mentions (7)

Other subjects in dontopedia point AT this entity as a value. These are inverse relationships — e.g. "X motherOf this subject" — and answer questions the forward facts can't. Grouped by predicate.

requiresRequires(2)

comparisonTargetComparison Target(1)

isTypeOfIs Type of(1)

performsNotMuchWorseThanPerforms Not Much Worse Than(1)

proposesSolutionProposes Solution(1)

relationshipToRelationship to(1)

Other facts (8)

The long tail: predicates that appear too rarely to warrant their own section. Filter or scroll to find a specific one. Each row links to its source.

8 facts
PredicateValueRef
Rdf:typeLlm[3]
Rdf:typeOriginal Model[6]
Rdf:typeModel Category[7]
Has Capability LevelSonnet Capability[1]
ProvidesMore Capacity to Learn Code to Byte Mapping[2]
SupportsFactual Recall Scaling[4]
Provides More Capacitytrue[5]
Characteristicbigger-than-bert[7]

Timeline

Timeline axis is valid_time — when each source says the fact was true in the world, not when Dontopedia learned about it. Retracted rows are kept for provenance; coloured stripes indicate the context kind.

9 undated
—
hasCapabilityLevelblah/general/part-128
ex:sonnet-capability
—
providesblah/watt-activation/part-292
ex:more-capacity-to-learn-code-to-byte-mapping
—
typeblah/general/127
ex:LLM
—
labelblah/watt-activation/178
much larger model
—
supportsblah/watt-activation/178
ex:factual-recall-scaling
—
providesMoreCapacityblah/watt-activation/290
true
—
typebeam/52a2411f-6cdc-40f7-817f-3feef46e4a6b
ex:OriginalModel
—
typebeam/63f3f6ff-b059-492e-954d-ccca67c2349d
ex:Model-category
—
characteristicbeam/63f3f6ff-b059-492e-954d-ccca67c2349d
bigger-than-bert

References (7)

7 references
  1. [1]Part 1281 fact
    ctx:discord/blah/general/part-128
  2. [2]Part 2921 fact
    ctx:discord/blah/watt-activation/part-292
  3. [3]1271 fact
    ctx:discord/blah/general/127
    • full textgeneral-127
      text/plain3 KBdoc:agent/general-127/dd0bc789-eaf1-449d-b955-e91c7e63e815
      Show excerpt
      [2026-04-07 12:49] ajaxdavis: someone make a full app/project benchmark where every step/layer, all models in competition exponential fan out and test all outcomes (lol bad description, here is a chatgpt generated diagram) (the point being …
  4. [4]1782 facts
    ctx:discord/blah/watt-activation/178
    • full textwatt-activation-178
      text/plain3 KBdoc:agent/watt-activation-178/50ec323b-637a-4c18-bba8-73839dc1355d
      Show excerpt
      [2026-03-10 00:29] xenonfun: ⏺ That settles it — the base instruct model has no factual knowledge at all. It generates fluent-sounding nonsense rather than facts. This is a pretraining depth problem: FineWeb-Edu teaches language patterns,…
  5. [5]2901 fact
    ctx:discord/blah/watt-activation/290
    • full textwatt-activation-290
      text/plain2 KBdoc:agent/watt-activation-290/22fb306d-bdbd-4a04-aa08-e5109c0026d8
      Show excerpt
      [2026-03-14 03:30] xenonfun: ⏺ Launched. The full pipeline (Phase 1 → encode → spherical Code LM → decoder → generate 15 samples) will take ~10 min. The Code LM now uses cos(h, w) / tau with tau=0.1 as the output head — matching the S^{d-1}…
  6. [6]52a2411f 6cdc 40f7 817f 3feef46e4a6b1 fact
    ctx:claims/beam/52a2411f-6cdc-40f7-817f-3feef46e4a6b
    • full textbeam-chunk
      text/plain1 KBdoc:beam/52a2411f-6cdc-40f7-817f-3feef46e4a6b
      Show excerpt
      - The model is pruned by removing 50% of the neurons in linear layers. This reduces the number of parameters and improves inference speed. 4. **Efficient Tokenizer**: - The `use_fast=True` option is used to enable the fast tokenizer …
  7. [7]63f3f6ff B059 492e 954d Ccca67c2349d2 facts
    ctx:claims/beam/63f3f6ff-b059-492e-954d-ccca67c2349d
    • full textbeam-chunk
      text/plain1020 Bdoc:beam/63f3f6ff-b059-492e-954d-ccca67c2349d
      Show excerpt
      However, I'm only achieving about 80% accuracy with this approach. I've studied LLM-based reformulation and noted a 25% intent accuracy boost for 6,000 complex queries. Can you help me improve my implementation to reach at least 92% detecti…

See also

Keep researching

Missing something or suspicious of what's here? Kick off a research session — a Claude agent will investigate, cite its sources, and file new facts into a dedicated context you can review before accepting into the shared view.

↑