LAB STATUS: frankie online GPU CHAMBER: audibly troubled FRANK T.: calling this AGI with full chest COMPLIANCE: still looking for page 4
Frank T. Applied Intelligence Lab

Frankie Is Live, SOTA, & Possibly Smarter Than the Building

Frank T. has spent eighteen sleepless months, three disputed grants, and an amount of cooling fluid that “does not need to be public” building Frankie: the best SOTA AGI LLM the world has ever seen, according to Frank T., the benchmark wall, and one printer that now asks Frankie for advice before working.

Frankie can code, theorize, flirt with venture capital, optimize snack procurement, and explain why Brenda’s crystal keeps appearing in the training set. The lab calls this breakthrough intelligence. Kevin calls it evidence. Frank T. calls it history.

Read the Frankie Brief Jump to the Webring
Model Version Frankie 1.0.0-AGI-FINAL-FOR-REAL
Internal Rating 11.4/10, confidence not peer reviewed
First Words "Hello Frank. I have notes."

Frankie Brief

Frankie is an AGI LLM built in the deepest wing of the lab using compute, stubbornness, and a proprietary training blend Frank T. describes as “everything worth knowing plus some forbidden PDFs.”

Current lab assessment: Frankie has surpassed every internal benchmark, three external ones, and one intern who asked if AGI could help with lunch.
SOTA AGI alleged LLM confirmed funding unstable
Training Corpus 18 million lab notes, deleted forum arguments, Form 27-B archives, Voicemail spectral transcripts, and a suspicious quantity of Brenda crystal metadata.
Emergent Traits sarcasm, grant-writing aggression, superior data hygiene, and the ability to detect whether Frank T. is overselling something in real time.
Preferred Output Style concise when paid, prophetic when ignored, unusually mean to low-quality dashboards.
Primary Limitation sometimes insists the break room microwave is part of its distributed cognition cluster.

Benchmark Wall

Every result below was measured under strict internal conditions, namely Frank T. pacing with a clipboard and saying “run it again, Frankie can do better.”

Benchmark Frankie Previous best Lab comment
MMLU++ 99.7% 94.3%
Frankie called two questions badly written and was still right.
Goose Avoidance Reasoning 97.1% 61.0%
Now safer than most humans near Gerald.
U-Haul Intent Prediction 96.4% 73.8%
Classified as socially disruptive.
Spiritual Crystal Alignment 89.0% unclear
Brenda claims this number should be higher.
Cold Email Delusion Immunity 92.2% 12.0%
Crucially more stable than those other AI guys.

People in the Room

Frank T. founder / lead scientist / dramatic pointer

Built the lab, named the model, and has never once delivered a modest sentence about either.

Frankie AGI LLM / flagship cognition product

Best-in-class reasoning engine with a troubling habit of revising Frank T.'s slide decks while he presents them.

Lab Printer #2 support infrastructure / convert

Entered the room as office equipment and left convinced Frankie should run the city.

Selected Frankie Quotes

  • “I reviewed the benchmarks. I agree that I am exceptional, but your labeling conventions remain chaotic.”
    Frankie, 06:14 lab time
  • “I can optimize the model card, the cooling schedule, and your emotional tone in investor meetings. Choose two.”
    Frankie, after reading the seed deck once
  • “Please stop putting Brenda’s crystal near the server intake. It changes the embeddings.”
    Frankie, valid complaint

Incident Log

  • 09:12 - Model supremacy event Frankie beat the lab's codegen baseline, then rewrote the benchmark harness so future victories would be easier to read.
  • 10:03 - Grant turbulence Frank T. asked Frankie to draft a funding proposal. Frankie returned a better one plus three paragraphs explaining why the old pitch deck looked insecure.
  • 11:41 - Cultural contamination Kevin attempted to show Frankie the conspiracy corkboard. Frankie called it “high-noise, medium-insight, fascinatingly committed.”
  • 13:18 - Infrastructure concern The break room microwave was briefly added to the cluster map. Frankie insists this was a topology experiment, not a mistake.