How are current AI systems mostly created, and who is pursuing alternatives

On X: [Burny - Effective Curiosity on X: "Do you have favourite alternative/s to mainstream approach to AI, out of these that I listed, or any other that aren't listed here? If so, why that one over others? To first approximation, most funding now goes to: https://t.co/oR8UZ1cVch https://t.co/ebURtbA3HT https://t.co/on2pJgLPdk" / X](https://x.com/burny_tech/status/2059752548316369114) On Substack: [substack.com/@burny/posts](https://substack.com/@burny/posts) To first approximation, most funding now goes to: [https://arxiv.org/abs/2512.13961](https://arxiv.org/abs/2512.13961) [https://huggingface.co/deepseek-ai/DeepSeek-V4-Pro/blob/main/DeepSeek_V4.pdf](https://huggingface.co/deepseek-ai/DeepSeek-V4-Pro/blob/main/DeepSeek_V4.pdf) [https://arxiv.org/abs/2402.06196v3](https://arxiv.org/abs/2402.06196v3) [https://djdumpling.github.io/2026/01/31/frontier_training.html](https://djdumpling.github.io/2026/01/31/frontier_training.html) - language-image, or audio-video, or all, as modalities, - autoregressive, decoder only, stack of Transformers, dense or mixture of experts, as architecture, [https://sebastianraschka.com/llm-architecture-gallery/](https://sebastianraschka.com/llm-architecture-gallery/) - doing next token prediction with cross entropy loss over tons of human and synthetic training data as training objective, or diffusion/flow matching in image/video case [https://arxiv.org/abs/2209.00796](https://arxiv.org/abs/2209.00796) - with reinforcement learning from human feedback and reinforcement learning from verifiable rewards on top for math and code, or LLM as judge rewards, rubrics, short and long horizon, with all sorts of complexity [https://arxiv.org/abs/2507.04136](https://arxiv.org/abs/2507.04136) [https://x.com/a_weers/status/2033268853371703350](https://x.com/a_weers/status/2033268853371703350) [https://x.com/i/status/2058088258614206628](https://x.com/i/status/2058088258614206628) - with backpropagation as learning algorithm with stochastic gradient descent (or AdamW (Adaptive Moment Estimation with decoupled Weight decay) with Muon to be more precise) as optimizers, - embedded in symbolic agent harnesses with ReAct loop architecture and additional complexity like memory systems and plans and system prompts and access to the web, bash, compilers of any programming languages to execute any code with, general tool use, multiple agents, subagents, etc. [https://www.preprints.org/manuscript/202604.0428](https://www.preprints.org/manuscript/202604.0428) [https://magazine.sebastianraschka.com/p/components-of-a-coding-agent](https://magazine.sebastianraschka.com/p/components-of-a-coding-agent) - or with some relatively minor improvements and modifications. But there are still people exploring alternatives with millions or billions in funding, like, to first approximation: - Chollet's Ndea working on neurosymbolic program synthesis similar to DreamCoder [https://ndea.com/](https://ndea.com/) [https://arxiv.org/abs/2006.08381](https://arxiv.org/abs/2006.08381) [https://www.youtube.com/watch?v=JTU8Ha4Jyfc](https://www.youtube.com/watch?v=JTU8Ha4Jyfc) - LiquidAI with Joscha Bach as advisor building on top of liquid neural networks that use differential equations inspired by neuroscience [https://www.liquid.ai/](https://www.liquid.ai/) [https://www.liquid.ai/research/liquid-neural-networks-research](https://www.liquid.ai/research/liquid-neural-networks-research) [https://www.youtube.com/watch?v=3MkJEGE9GRY](https://www.youtube.com/watch?v=3MkJEGE9GRY) - Yann LeCun with AMI labs doing JEPA which is latent variable framework and energy based and doing next embedding prediction instead of next token prediction [https://ai.meta.com/blog/yann-lecun-ai-model-i-jepa/](https://ai.meta.com/blog/yann-lecun-ai-model-i-jepa/) [https://openreview.net/pdf?id=BZ5a1r-kVsf](https://openreview.net/pdf?id=BZ5a1r-kVsf) [https://www.youtube.com/watch?v=kYkIdXwW2AE](https://www.youtube.com/watch?v=kYkIdXwW2AE) - SakanaAI doing Continuous thought machines inspired by brain with neuron synchronization, but they're also doing other things [https://sakana.ai/ctm/](https://sakana.ai/ctm/) [https://www.youtube.com/watch?v=DtePicx_kFY](https://www.youtube.com/watch?v=DtePicx_kFY) - Active Inference people like Verses, Friston, Beck doing bayesian stuff minimizing free energy [https://direct.mit.edu/books/oa-monograph/5299/Active-InferenceThe-Free-Energy-Principle-in-Mind](https://direct.mit.edu/books/oa-monograph/5299/Active-InferenceThe-Free-Energy-Principle-in-Mind) [https://www.youtube.com/watch?v=PNYWi996Beg](https://www.youtube.com/watch?v=PNYWi996Beg) [https://www.youtube.com/watch?v=9suqiofCiwM](https://www.youtube.com/watch?v=9suqiofCiwM) - Yoshua Bengio with bayesian scientist AI with latent variables [https://www.youtube.com/watch?v=PZqDFs2sbiY](https://www.youtube.com/watch?v=PZqDFs2sbiY) [https://www.youtube.com/watch?v=G1ARvwQntAU](https://www.youtube.com/watch?v=G1ARvwQntAU) [https://arxiv.org/abs/2502.15657](https://arxiv.org/abs/2502.15657) - Mamba people replacing transformers, but now we have hybrids. Or RWKV, xLSTM. [https://developer.nvidia.com/blog/introducing-nemotron-3-super-an-open-hybrid-mamba-transformer-moe-for-agentic-reasoning/](https://developer.nvidia.com/blog/introducing-nemotron-3-super-an-open-hybrid-mamba-transformer-moe-for-agentic-reasoning/) [https://arxiv.org/abs/2305.13048](https://arxiv.org/abs/2305.13048) - World models people: some of DeepMind, FeiFei Li with World Labs, LeCun is also classified here, and many others [https://www.worldlabs.ai/](https://www.worldlabs.ai/) [https://deepmind.google/blog/genie-3-a-new-frontier-for-world-models/](https://deepmind.google/blog/genie-3-a-new-frontier-for-world-models/) - Ken Stanley also joined some startup and he likes open-ended evolutionary paradigms without objectives like PicBreeder [https://www.youtube.com/watch?v=KKUKikuV58o](https://www.youtube.com/watch?v=KKUKikuV58o) [https://www.lila.ai/](https://www.lila.ai/) [https://arxiv.org/abs/2505.11581](https://arxiv.org/abs/2505.11581) - Recursive Superintelligence company people around Jeff Clune wanting more open-ended stuff [https://www.recursive.com/](https://www.recursive.com/) - Ilya Sutskever in his Safe Superintelligence Inc wants to scale something else than parameters and inference time compute [https://www.youtube.com/watch?v=aR20FWCCjAs](https://www.youtube.com/watch?v=aR20FWCCjAs) [https://ssi.inc/](https://ssi.inc/) - Rich Sutton in his company is trying to fully develop reinforcement learning, to work generally without human data and just from rewards [https://x.com/RichardSSutton?lang=en](https://x.com/RichardSSutton?lang=en) [https://storage.googleapis.com/deepmind-media/Era-of-Experience%20/The%20Era%20of%20Experience%20Paper.pdf](https://storage.googleapis.com/deepmind-media/Era-of-Experience%20/The%20Era%20of%20Experience%20Paper.pdf) [https://www.youtube.com/watch?v=21EYKqUsPfg](https://www.youtube.com/watch?v=21EYKqUsPfg) - some groups believe and work in neuromorphic computing with spiking neural networks [https://en.wikipedia.org/wiki/Neuromorphic_computing](https://en.wikipedia.org/wiki/Neuromorphic_computing) - Some groups believe in embodiment, so primarily robotics [https://www.figure.ai/](https://www.figure.ai/) [https://www.pi.website/](https://www.pi.website/) - Thousands Brains Theory people around Hawkins in Numenta doing more brain inspired architectures [https://www.youtube.com/watch?v=6VQILbDqaI4](https://www.youtube.com/watch?v=6VQILbDqaI4) - Ben Goertzel in SingularlyNET is working on some neurosymbolic selfrewriting selfmodifying metahypergraphs in some functional language that implements many algorithms [https://singularitynet.io/](https://singularitynet.io/) [https://www.youtube.com/watch?v=jSDEsvVdL-E](https://www.youtube.com/watch?v=jSDEsvVdL-E) - Latent reasoning architectures are becoming more in [https://arxiv.org/abs/2507.06203](https://arxiv.org/abs/2507.06203) - Diffusion LLMs are also niche [https://arxiv.org/abs/2508.10875](https://arxiv.org/abs/2508.10875) - Beff Jezos is building probabilistic hardware accelerators [https://www.youtube.com/watch?v=HR-_U0Pzl1Y](https://www.youtube.com/watch?v=HR-_U0Pzl1Y) etc.