current AI companies are mostly going to pivot to really dystopian commercial use cases to try to get some of their money back. romantic chat bots, targeted ads, surveillance/data selling, etc

(http://www.autoadmit.com/thread.php?thread_id=5734039&forum_id=2).#48991123)

even if the limit of current AI is automating much of white collar knowledge work (programming, corporate law, finance, etc.) that's still a very large and fundamental shift. there's also stuff like having an interactive personal tutor which can adapt to your exact skill level/knowledge available on demand. multimodal models being the "missing piece" to make general purpose robots feasible, even if their use is very limited at first. etc.

meta has also been an absolute fucking disaster in the AI race given the amount of resources they've put into it

(http://www.autoadmit.com/thread.php?thread_id=5734039&forum_id=2).#48992167)

everyone smart has figured out that "AGI" is not coming from LLMs. makes that AI 2027 report really funny in retrospect. maybe people should stop listening to scott alexander and miscellaneous indians? also ljl at yudkowsky et al shamelessly continuing their doomerism grift

(http://www.autoadmit.com/thread.php?thread_id=5734039&forum_id=2).#48992175)

for first world users, yeah, targeted ads will be very profitable, and all of the frontier model companies will do them

(http://www.autoadmit.com/thread.php?thread_id=5734039&forum_id=2).#48992197)

this whole interview is quite good tbh

(http://www.autoadmit.com/thread.php?thread_id=5734039&forum_id=2).#48992216)

i believe that privately, LLM proponents know that this is complete nonsense, and we're nowhere even remotely near AGI, and it's not coming from LLMs. but publicly, people are still taking these delusions seriously, because it's to their advantage to do so in order to hype up their own self-interest

(http://www.autoadmit.com/thread.php?thread_id=5734039&forum_id=2).#48992250)

it's interesting to note that that while certain things with easily verifiable rewards are advancing rapidly still (such as programming and math benchmark performance), other things seem to be stalling. the Claude 4 benchmarks don't see too promising for the LLM maximalist point of view but it's hard to know and we only have limited data points. the other issue is that we have very little idea what the labs are working on internally. transformers with chain of thought might soon hit a wall, but how can we be confident that someone doesn't have something else cooking that could address the remaining problems? LLMs are not necessarily synonymous with transformers trained with stochastic gradient descent.

(http://www.autoadmit.com/thread.php?thread_id=5734039&forum_id=2).#48992303)

it's pretty hard, man. that's a huge leap from what we have right now. the recent models that have been released across the board have been very small improvements and it's very clear that progress has massively slowed down and we're not going to see the same scaling off added compute trendline going forward

(http://www.autoadmit.com/thread.php?thread_id=5734039&forum_id=2).#48992410)

1. Core Idea: Predicting in Latent Space

Traditional self-supervised approaches—like autoencoders or generative masked‐modeling—often try to reconstruct pixels or raw tokens, which forces the model to spend capacity on both relevant and irrelevant details (e.g., exact pixel colors). JEPA sidesteps this by having two networks:

A context encoder that processes observed parts of the input (e.g., an image with masked regions or a video clip missing certain frames) and produces a “context embedding.”

A target encoder that separately encodes the actual data (or future frames) into “target embeddings.”

The training objective is to align the context embedding with the correct target embedding (or to distinguish it from incorrect ones) in latent space, rather than to reconstruct raw pixels or tokens. By comparing embeddings directly, the model can discard unpredictable noise (e.g., lighting variations, background clutter) and focus on stable, high-level features that are useful for prediction and planning

turingpost.com

arxiv.org

.

2. Architecture Variants (I-JEPA, V-JEPA, etc.)

I-JEPA (Image JEPA): Given a single image, a large “context crop” (covering a broad spatial area) is encoded, and the model predicts embeddings of several “target crops” from that image. Target crops are often chosen at scales large enough to require understanding semantics (e.g., object identities), not trivial low-level details

ai.meta.com

.

V-JEPA (Video JEPA): Extends I-JEPA to video by having the context encoder ingest previous frames (and possibly actions), then predicting the embedding of future frames. Because it only needs to predict abstract representations, the model can choose which features of the future are predictable (e.g., object positions) and ignore the unpredictable (e.g., exact pixel noise)

linkedin.com

ai.meta.com

.

By operating in this “embedding space” rather than pixel space, JEPA-based models learn world‐model representations: latent features that capture how a scene or environment evolves over time (e.g., object motion, physical interactions) without being burdened by pixel-level reconstruction

arxiv.org

medium.com

.

3. Loss Function and Training Dynamics

JEPA typically uses a contrastive or predictive loss at the embedding level. A common choice is InfoNCE: the context embedding must be close (in representation space) to the true target embedding and far from negative samples (embeddings of unrelated patches or frames). In some variants, an exponential moving average is used to stabilize the target encoder, ensuring that the targets change more slowly than the context encoder (similar to BYOL or MoCo strategies)

arxiv.org

.

Because the model is encouraged to predict only abstracted features, it effectively learns which aspects of the environment are predictable and worth modeling. For instance, in V-JEPA, predicting where a car will be next frame is feasible, whereas predicting the precise noise pattern on its surface is not. By focusing capacity on the predictable latent variables, JEPA induces a more robust internal “world model” that can be reused for downstream tasks (classification, reinforcement learning, planning) with far fewer labeled samples

linkedin.com

arxiv.org

.

4. Why JEPA Enables Self-Formed World-Models

Abstract Prediction vs. Generative Modeling: Generative models (e.g., diffusion, autoregressive transformers) must allocate capacity to model every detail, including inherently unpredictable factors. JEPA’s abstraction means that if some aspect of the future cannot be predicted from the context (e.g., random background flicker), the model can “discard” it and focus on the stable dynamics (e.g., object trajectories)

ai.meta.com

arxiv.org

.

Efficiency & Generalization: Empirically, JEPA variants (I-JEPA, V-JEPA) show 1.5×–6× gains in sample efficiency compared to pixel-based generative pre-training, because they aren’t forced to learn noise patterns or outliers. This leads to embeddings that capture “common sense” world dynamics—e.g., gravity, object permanence—encouraging the model to form its own latent simulation or predictive engine that generalizes to new tasks with minimal adaptation

linkedin.com

medium.com

.

Scalability & Modularity: The separation between context encoder and target encoder (or predictor) means that JEPA can be stacked hierarchically. A higher-level JEPA might predict scene-level embeddings (e.g., “a red car turns right”), while a lower-level JEPA predicts pixel embeddings or optical flow. This hierarchy mirrors how humans build world models: first conceptualizing objects and actions, then filling in details

rohitbandaru.github.io

medium.com

.

5. Practical Outcomes & Extensions

Recent work has shown that JEPA-trained backbones (e.g., ViT with I-JEPA) outperform standard self-supervised baselines on tasks like object detection, depth estimation, and policy learning when used as initializations

arxiv.org

. Furthermore, extensions like seq-JEPA incorporate sequences of views plus “action embeddings,” allowing the model to learn representations that are both invariant (for classification) and equivariant (for tasks requiring precise spatial dynamics), effectively learning a richer world model in a single architecture

arxiv.org

.

In summary, JEPA’s strength lies in its ability to force the model to abstract away unpredictable noise and extract only the predictable, semantically meaningful features of its inputs. By learning to align context embeddings with the embeddings of masked or future data, the model inherently constructs an internal world model—a latent simulation of its environment—that can be leveraged for downstream reasoning, planning, and decision-making with high efficiency.

(http://www.autoadmit.com/thread.php?thread_id=5734039&forum_id=2).#48992758)