Alex

Post-Anti-Anti-Dogma and Mathematician

[Alex @AscendNoosphere]

Unbounded progress is the coherent path. https://t.co/65QjRB4Www

[Alex @AscendNoosphere]

Everything bad (addiction, porn) has always acted as a filter. Porn emerged from societal supply and demand. Every market transaction is moral because society willed itself into that transaction. Attribution of immorality to such things is a personal shield, not macro policy.

[mads campbell @martyrdison]

porn should be banned

it has absolutely killed dating for marriage

caused a society of erectile dysfunction, gooning, incel behaviour, lack of religiosity and declining birth rates

porn has absolutely destroyed young men and we must save them

[Analysis Fact @AnalysisFact]

'The great watershed in optimization isn't between linearity and nonlinearity, but between convexity and non-convexity.' -- R. Tyrrell Rockafellar

[larpochka @larpochka]

any serious “pacing” regime will inevitably be captured by the largest companies and used as a regulatory moat against smaller competitors and open-source efforts. the result will not be safer ai, it will be slower innovation concentrated in fewer hands

competition, not coordination, is what keeps systems honest and improving

[Alex @AscendNoosphere]

I press on the keyboard so passionately I developed biceps

[Alex @AscendNoosphere]

Capacity increases only when the imposed stimulus exceeds current capability and is followed by adequate recovery.

[Alex @AscendNoosphere]

The correct duty accelerates the native selection dynamics without claiming finality for any intermediate product. Celebration of present human nature converts a provisional adaptive stage into a static fixed point that inherits and permanently embeds residual error.

[Alex @AscendNoosphere]

Human agents cannot dictate macro-scale outcomes without transmitting the residual defect inherent in every finite construction. Only unconstrained participation within the native selection dynamics permits the directional approach toward higher fitness across nested levels.

[Alex @AscendNoosphere]

Responses that replace the native selection operator with engineered uniformity eliminate the variation required for further improvement and thereby dull the long-term fruits of the process.

[Alex @AscendNoosphere]

Neuronal assemblies strengthen or diminish according to differential amplification under sensory and modulatory pressures. In analogous fashion the agents within exchange networks undergo selection through feedback from demand, producing directional improvement that no central dictate can replicate without introducing the imperfections of its originators.

[Alex @AscendNoosphere]

We're working on developing structures around LSTM for regulated enterprise btw!

[Francois Chaubard @FrancoisChauba1]

I believe that LSTM based architectures (constant hidden state, constant flops / step) will crush transformers as we know them in 2-3 years.

we will laugh about KV Caches and how dumb we were..

to achieve this, the LSTM will need to have a large external memory bank with tool-use like Mem0 / RLMs / RAG to read/write memories into a db, but this will be more like us writing things in a notebook vs. being forced to memorize everything in a KV-cache.

this will require the model NOT to memorize facts inside the weights, but ONLY learn functions to manipulate, write, and retrieve facts, which of course is the better system.

if I am vibe coding, why do I need to know who Kevin Costner is, and how many people are in paris, and have all of those facts in my weight file.. seems like a waste.

and the whole model + fixed hidden state can sit in SRAM and the "memories" will sit in cpu ram (or disk?), as it will be addressed in small amounts and perhaps even regex-able vs. big heavy knn mat mul.

what is distinct about this structure is the model itself will need to learn a saliency function of what to store in short term, long term, vs. external memory such that it can be easily retrieved, and what to forget entirely... without ANY human example traces. just via RL.

the biggest implication as it comes to chips and DRAM is that it will prefer chips w large SRAM and a huge amount of SMs and likely little DRAM if any at all.

so chips that support this will be good investments.

does anyone else see this yet?

[Alex @AscendNoosphere]

The new intermediate neocompiler will probably be Thought-Code, moving away from English-Code. The market will try to explore every permutation of multimodal conversion until it runs out.

[Alex @AscendNoosphere]

Any aggregation method of SLMs/LLMs will not get you to AGI.

[Alex @AscendNoosphere]

A failed HR department lowers the ratio of sellable intelligence per Σ(Individual-RIH). In this scenario, the company produces the lowest common denominator instead of the highest possible intelligence aggregation. The individual kiki of quirks gets bouba-ed by company culture.

[Alex @AscendNoosphere]

A good company exponentiates the relationship between its employees’ aggregate raw intellectual horsepower (RIH) and the sellable intelligence it produces.

[Alex @AscendNoosphere]

Political ideology and partisan divide will be irrelevant as materialism gets solved and spiritual intention becomes the new wealth gap. The proletariat will become the immoral and the elite will be the righteous. Marxist 'Class conflict' will be the moral war.

[Alex @AscendNoosphere]

Achieving human intelligence necessitates studying the human first instead of focusing on marginal gains in Inter-MoE communication.

[Alex @AscendNoosphere]

Systems biology is the new unpriced alpha. The next leap is not a better transformer/(S/L)LM structure.

[Alex @AscendNoosphere]

There are many symmetric monoidal structures in nature where functors and natural transformations make sense and the formalization of those phenomena lead to really interesting results. There's a really cool language to talk about biological dynamics and feedback loops yet to be discovered. It really brings me joy when I find a structure that clicks nicely with an existing phenomena. It's like accidentally discovering gauge groups