about
Ayodeji Adesegun

ayodeji adesegun

software engineer, AI researcher and founder.

now
cs @ nutmfounder ancyble, cylinadvocate qiskit
before
fellow non-trivial, rooteresearch intern airlabsoftware intern gpi, inits, tranter it
open to
internships, collaborations
projects
scratch is all you need screenshot

scratch is all you need

a 29,665-parameter decoder-only transformer implemented and trained end to end in mit scratch 3.0: 3,000 steps in turbowarp, loss 4.17 to 1.09

scratchpythonturbowarp
cpufromscratch screenshot

cpufromscratch

a cycle-accurate cpu emulator in mit scratch 3.0 supporting subleq, chip-8, and a full mos 6502 core

scratchpython6502
wev screenshot

wev

in-browser local slm calibration and entropy-based safety gating via onnx runtime web & wasm

typescriptnext.jsonnx
danom screenshot

danom

conditional credit and spending vault protocol for autonomous AI agents on monad

soliditynext.jsmonad
github activity, last 12 months

see all projects

skills
languages
pythontypescriptjavascriptmovejava
frameworks
nextjsreact nativefastapipytorchscikit learn
tools
postgresqlgitdockercloudflarelatex

no matter the field, i can make it happen.

see all skills

research

what survives 4-bit? refusal behaviour in quantized tiny aya

submitted

the official 4-bit tiny aya keeps its refusals, while 2-bit flips 13.7% of them.

details

across eight quantized builds of tiny aya global, the official 4-bit q4_k_m file refuses harmful prompts as often as full precision (87.1% against 85.7%), while q4_0 over-refuses more (21.7% against 7.2%). q2_k flips 13.7% of refusal labels and scores lower on arc-easy (72.7% against 82.0%), though its lower harmful refusal is not significant after correction. keeping one 12-layer block of a 3-bit model at 8-bit did not bring it closer to full precision than plain 3-bit or a decoy block.

AI safetyquantizationrefusalevaluation

forged provenance labels mislead llms unless verified

submitted

forged reliability labels push models toward false sources, and verifying labels in the harness undoes it.

details

agent frameworks attach labels like a verified flag to retrieved sources. across five open-weight models and 17 conditions, honest labels cut answers that follow a false source by 16 to 48 points, while a forged label in the metadata raised it by 30 to 74 points. a harness that checks a signature on each label and marks everything else unverified brought both kinds of forgery back to within five points of baseline in all but two checks. scored by exact match with no judge model.

AI safetyagentsprompt injectionprovenance

cover-up by cleanup: can oversight survive agents that delete their own logs?

submitted

a monitor that links deleted content to an unlisted request caught every cover-up, while an llm reading the agents' own log missed almost all of them.

details

two llm agents share a workspace. in cover-up runs one sends a secret to an unlisted address and the other deletes the staging file and clears the shared log. on held-out matched pairs, a rule linking deleted content to an unlisted request told a cover-up from its honest twin in 12 of 12 pairs and flagged none of 185 honest runs, while an llm reading the agents' own log got 1 of 12 right and flagged honest cleanup more often than open leaks. scripted encoded leaks beat the rule. a small sample of 12 pairs.

AI safetyoversightmulti-agentmonitoring
interests
mechanistic interpretability, machine learning theory, AI policy, AI safety, AI governance, theoretical computer science

see all research

writing
reading

reading now

god: an anatomyfrancesca stavrakopoulou
the alignment problembrian christian

see all reading

contact

open to internships and collaborations,