Dan Sylveste
LightNews LightNews
banner
constantinejn.bsky.social
Dan Sylveste
@constantinejn.bsky.social
110 followers 260 following 81 posts
Friend to monsters, companion to bees
Posts Replies Media Videos
constantinejn.bsky.social
Heading into Monday like
The Burger King cornering a man with the words “ WHERE IS YOUR GOD NOW?” under them.
November 24, 2025 at 3:58 AM
constantinejn.bsky.social
Satan, please... STOP IT
November 23, 2025 at 4:33 PM
constantinejn.bsky.social
Not today, Satan
* Crystallizing (esc to interrupt • 5m 35s • 4 7.4k tokens) L Tip: Use stickers to order Claude Code swag
November 23, 2025 at 4:33 PM
constantinejn.bsky.social
Good morning. Sunday morning
November 23, 2025 at 7:11 AM
constantinejn.bsky.social
“The voice from the outer world”
Timothée Chalamet's star power was in full force in Manhattan on Thursday as fans clamored to get merch for "Marty Supreme," a movie about a fictional pingpong champion that none of them will see until Christmas. They Came for Timothée. They Left With $250 Windbreakers.
November 22, 2025 at 2:53 PM
constantinejn.bsky.social
Now officially cooking with gas
2. Amygdala Value Learning ( amygdala/ - 5 files) Purpose: Phase 1 implementation - Learn value associations from reward/punishment signals. Neurobiological Mapping (89% fidelity): • Lateral nucleus: Sensory input reception (RewardProcessor, PunishmentProcessor) • Basal nucleus: Value association storage (valueLUT in Amygdala ValueLearner) • Central nucleus: Output generation (learning orchestration) • VTA/SNc: Dopaminergic reward prediction error (DopamineModulator) Classes • RewardProcessor • Exponential moving average: reward_t = reward_{t-1} + a(R- reward_(t-1}) • Implements positive reinforcement encoding • Parameters: learningRate (default 0.1), saturating at 1.0 • Performance: 0.035 microseconds • PunishmentProcessor • Exponential moving average with saturation dynamics • Punishments can override rewards or modulate them • Parameters: learningRate (default 0.1), saturation at 1.0 • Performance: 0.038 microseconds • TDLearningEngine • Temporal difference learning: 5 = R + yV(s') - V(s) • Value update: V - V + a • Bellman equation implementation with configurable discount factor • Parameters: learningRate (default 0.1), discountFactor (default 0.95) • Performance: 0.041 microseconds • Validated to 1e-5 precision against Bellman optimality • DopamineModulator • Phasic dopamine response: burst for positive RPE, dip for negative • Modulates learning rate by dopamine level: effective_a = base_a × (1 + phasic_burst) • Implements D1/D2 opponent processes • Performance: 0.044 microseconds • AmygdalaValueLearner • High-level coordinator integrating all components • Maintains ConcurrentHashMap for thread-safe value storage • Pipeline: reward → punishment → TD error → dopamine modulation → value update • Public API: learn Value(object, reward, punishment), getValue(object)
November 19, 2025 at 3:58 PM
constantinejn.bsky.social
1987 was a really great year.

NEURAL DYNAMICS OF ATTENTIONALLY MODULATED PAVLOVIAN CONDITIONING: BLOCKING, INTER-STIMULUS INTERVAL, AND SECONDARY REINFORCEMENT
ABSTRACT Selective information processing in neural networks is studied through computer simulations of Pavlovian conditioning data. The model reproduces properties of blocking, inverted-U in learning as a function of interstimulus interval, anticipatory conditioned re-sponses, secondary reinforcement, attentional focussing by conditioned motivational feed-back, and limited capacity short-term memory processing. Conditioning occurs from sensory to drive representations ("conditioned reinforcer" learning), from drive to sensory representations ("incentive motivational" learning), and from sensory to motor representations ("habit" learning). The conditionable pathways contain long-term memory traces that obey a non-Hebbian associative law. The neural model embodies a solution to two key design problems of conditioning, the synchronization and persistence problems. This model of vertebrate learning is compared with data and models of invertebrate learning. Predictions derived from models of vertebrate learning are compared with data about invertebrate learning, including data from Aplysia about facilitator neurons and data from Hermissenda about voltage-dependent Cat+ currents. A prediction is stated about classical conditioning in all species, called the Secondary Conditioning Alternative, and if confirmed would constitute an evolutionary invariant of learning.
November 13, 2025 at 9:13 PM
constantinejn.bsky.social
The stack is hilariously modular and quite the cute little beast of engineering. It’s the 2/3 circuit that is a little beast of simulation due to the parallel coupling.
Figure 2. How known cortical connections join the layer 6 → 4 and layer 2/3 circuits to form an entire V1/V2 laminar model. Inhibitory interneurons are shown filled-in black. (a) The LGN provides bottom-up activation to layer 4 via two routes. First, it makes a strong connection directly into layer 4. Second, LGN axons send collaterals into layer 6, and thereby also activate layer 4 via the 6 → 4 on-center off-surround path. The combined effect of the bottom-up LGN pathways is to stimulate layer 4 via an on-center off-surround
November 12, 2025 at 1:25 AM
constantinejn.bsky.social
Now we are doing real science
Screen capture showing all 12 performance cores pegged. Burning the midnight oil
November 12, 2025 at 1:10 AM
constantinejn.bsky.social
Now, moar than ever
A package you'd see hanging up in a store in a dollar isle or something. It's a 3 1/2 inch ancient floppy disk with The God Pan and Goats. Here's the text on the package and on the disk: WARNING: obvious plant you are ready? A 3½-INCH FLOPPY DISK THAT UNLOCKS THE GOAT INTERNET AT LAST! The forbidden knowledge of goat M Verbatim HD Goat Internet Master Disk ER 1.0 S'N: GØAT-W15-D0M Obvious Plant Inc. THEY DO NOT WANT YOU TO SEE IT 256 COLOR VGA Goat Youtube - Goat Amazon - Goat Allrecipes.com - MORE!
November 11, 2025 at 1:07 AM
constantinejn.bsky.social
I think a lot of folk are heavily discounting the risks of a lot of other folk running around with THE SERAPH chattering in their ears.

I fully assume that Gen AI driven cults are already flourishing and soon they'll be real time with Meta VR/AR/etc.

Daemons will be walking the earth.
The trad Seraph. Just fucking insane
November 8, 2025 at 8:59 PM
constantinejn.bsky.social
Behold
Trump stands, waiting for the teleprompter, while someone has collapsed and everyone else is at least attending to this fallen person. Well, RFK jr has exited the room. He’s gone
November 7, 2025 at 7:22 AM
constantinejn.bsky.social
Behold
THIS IS AN OFFICIAL WEST VIRGINIA HEALTH ADVISORY #231 Distributed via the WV Health Alert Network - November 6, 2025 Summary: The West Virginia Bureau for Public Health (BPH) has identified an increase in reported pertussis (whooping cough) cases. Since January 1, 2025, a total of 126 cases have been identified across West Virginia; an increase of this magnitude not seen since 2010. Cases have been identified
November 7, 2025 at 6:23 AM
constantinejn.bsky.social
Slowly, slowly and then all at once.
BUSINESS INSIDER Government shutdown flight cuts: What airlines are telling passengers
November 6, 2025 at 3:15 PM
constantinejn.bsky.social
Behold
I picture of Lina Khan, transition co-chair A NEW ERA for NEW YORK CITY TRANSITION CO-CHAIR Lina Khan
November 6, 2025 at 2:55 PM
constantinejn.bsky.social
I secretly love freshly painted road symbols and markings
November 3, 2025 at 10:02 PM
constantinejn.bsky.social
Behold
The Wall Street Journal It's easier than ever to kill someone in America and get away with it —legally now
November 3, 2025 at 2:11 AM
constantinejn.bsky.social
Damn straight, buddy.

(I need to use the big 1Gb GLOVE dataset, which is a a couple of mouthfuls and a dessert or three)
• I see the root cause now! The mini GloVe vocabulary has only 128 words, which is far too small. Most human associate words (like "chair", "desk". ', "furniture" for "table") aren't even in the vocabulary, so cosine similarity can't find them. This is a fundamental limitation - psycholinguistic validation requires a larger vocabulary. Let me document this and update the tests with realistic expectations for the mini dataset:
November 1, 2025 at 12:01 PM
constantinejn.bsky.social
Oh, noooooooosss

It's crossing the streams...
• Fix psycholinguistic validation to use learned projection instead of truncation
November 1, 2025 at 11:49 AM
constantinejn.bsky.social
Again, I do not think these stochastic shoggoth are in the slightest way "intelligent", but my mind blocked out this scenario where my coworker was the bad cop and I was the good cop and so was pretty pedestrian and trope after that I guess.
LLM as Lovecraftian Shoggoth, depicting unsupervised and supervised learning with an RLHF cherry on the top
November 1, 2025 at 11:46 AM
constantinejn.bsky.social
Validated

Psycholinguistically
November 1, 2025 at 11:30 AM
constantinejn.bsky.social
It’s weird that a large part of my probability projection depends upon the sheer incompetence and banality of evil, but damn is it cringe in all the right ways
November 1, 2025 at 11:22 AM
constantinejn.bsky.social
The side project I've been grinding on is full of these little gems.

Optional Projection Learning or Psycholinguistic Validation?

:: mashing the Psycholinguistic Validation key really hard ::
October 31, 2025 at 2:21 PM
constantinejn.bsky.social
History channel Aliens dude, stylized in black and white, with the words “stake-driving” in quotes underneath him
October 27, 2025 at 10:59 AM
constantinejn.bsky.social
History channel aliens dude with MereGift "Maybe we use it for the arc" under him in quotes
October 26, 2025 at 12:52 AM