Amabwiriza ya AI

Impamvu Video ya AI ibona physics nabi

Videwo za AI zikora amakosa mu bumenyi bw'imbaraga (physics) kuko ibyitegererezo byiga imiterere y'imibare y'uko utudomo (pixels) akenshi duhinduka uko igihe gihita, aho kwiga amategeko y'uburemere, imbaraga n'igongano.

Umva iki gitaboAI-byavuzwe · ~3 min

Kuriyi page4 min soma
  1. Incamake
  2. Kwibira cyane
  3. Gushyira mu bikorwa Isi
  4. The Future of Why AI Video Gets Physics Wrong
  5. Kubishyira mubikorwa
  6. Komeza Ubushakashatsi
  7. Ibibazo bikunze kubazwa

Incamake

They imitate the look of motion without simulating what causes it. That is why objects vanish, liquids act strangely, limbs pass through things and clips stay short, and it matters for anyone trying to judge whether footage is real or use these models for planning and robotics.

Kwibira cyane

Video generators are trained on huge collections of footage to predict or denoise frames. The training objective rewards output that resembles real video. It does not reward output that obeys conservation of mass or momentum. Common physics, such as a ball falling or a person walking, is well represented in the data and usually looks right. Rare or complex events, such as glass shattering, a specific collision, or liquid pouring into an oddly shaped container, appear less often and are harder to learn. OpenAI's own technical material on Sora noted that it did not accurately model interactions like glass shattering. Object permanence is a memory problem. When something is hidden, the model has to carry its identity through frames where it is invisible. Models attend over a limited amount of context, so hidden objects can come back changed or not at all. Cause and effect is also weak: a bite, a spill or a dent should leave a lasting change, but the model may treat it as a passing visual event. Clip length is limited mainly by compute. A video is represented as spacetime tokens, and attention cost rises steeply as the token count grows, so many products generate seconds rather than minutes. Longer videos are often made by extending clips, which lets errors compound. A common misconception is that these models contain a physics engine. They do not, though they may pick up some implicit physical regularities. Researchers have built benchmarks such as Physics-IQ to test this and found that visually realistic output does not mean physical understanding. World-model research tries to close the gap: Google DeepMind's Genie line generates interactive environments, Meta's V-JEPA learns by predicting in an abstract representation space instead of pixels, and NVIDIA's Cosmos targets physical AI such as robotics.

Gushyira mu bikorwa Isi

A generated person bites a cookie, but afterward the cookie has no bite mark. OpenAI named this kind of failure among Sora's limitations when it previewed the model in 2024.

A dog walks behind a tree and comes out the other side with a different coat pattern, because the model did not keep a stable memory of the hidden object.

A runner's legs appear to swap sides mid-stride and the feet slide along the ground, which shows the model learned the look of running, not the contact forces involved.

A basketball passes through the rim, briefly vanishes, and reappears in a player's hands, a sequence that looks smooth frame to frame but makes no physical sense.

The Future of Why AI Video Gets Physics Wrong

Physical plausibility has improved with scale and better data, and it is likely to keep improving for common scenes. Harder cases, such as long-horizon cause and effect, rare interactions and reliable object permanence across long occlusions, may need architectural changes like explicit memory, 3D structure or hybrid simulation rather than scale alone. World models aimed at robotics and interactive environments are an active research area, and benchmarks that separately test visual realism and physical correctness will help show real progress. It is not settled whether pure video prediction can learn robust physics, and researchers disagree about it.

Kubishyira mubikorwa

Impamvu ari ngombwa

  • Umuvuduko n'igipimo. AI igaragara irashobora gukora igenzura, gutahura, no gutondekanya imirimo kurwego.
  • Kubaka amahitamo. Amakipe arema arashobora prototype ibitekerezo byihuse hamwe nintoki nkeya.
  • Itsinda hamwe nakazi. Ibikorwa birashobora gukoresha amashusho nibimenyetso bya videwo byari bigoye gutunganya.

Witondere

  • Uburenganzira bwishusho hamwe no kwemererwa birashobora guhinduka ibyago byemewe n'amategeko niba ibimenyetso bidasobanutse.
  • Imikorere yicyitegererezo irashobora gutandukana kumurika, demografiya, nibidukikije.
  • Ibyiza byibinyoma birashobora kutamenyekana keretse niba ibyiringiro byateganijwe bikurikiranwa.

Intambwe ikurikira

  1. Sobanura ibipimo byo kwemererwa kugiciro, kwibutsa, nibiciro byamakosa.
  2. Gerageza hamwe namakuru ajyanye nuburyo nyabwo bwo gukora.
  3. Ongeraho isubiramo ryabantu kubwizere buke cyangwa guhanura cyane.
  4. Kurikirana icyitegererezo cya drift hanyuma uhindurwe nyuma ya kamera cyangwa dataset ihinduka.

Komeza Ubushakashatsi

Ibaruwa y'amakuru y'ubuntu

Ibuka amakuru ya buri munsi ya AI

Inkuru eshatu zizewe za AI buri gitondo cy'iminsi y'akazi, zanditse mu rurimi rworoshye. Ubuntu iteka ryose, nta kwamamaza.

Imeri imwe buri munsi y'akazi. Kwiyimura mu buryo bumwe. Ntitwigera tugurisha cyangwa ngo dusangire aderesi yawe.

Isuzume

Take the Why AI Video Gets Physics Wrong quiz

Ibisubizo ako kanya kuri buri kibazo, n'icyemezo gishobora gusangirwa gifite ID yagenzurwa umaze gutsinda isomo.

Tangira ikibazo

Fasha uburezi bwa AI bw'ubuntu. AI Understanding ni umuryango utegamiye kuri leta wa 501(c)(3) — nta kwamamaza, nta kwishyura, iteka ryose. Tanga impano

Ibibazo bikunze kubazwa

What is Why AI Video Gets Physics Wrong?

AI video gets physics wrong because generators learn statistical patterns of how pixels usually change over time, not the rules of mass, force and collision. They imitate the look of motion without simulating what causes it. That is why objects vanish, liquids act strangely, limbs pass through things and clips stay short, and it matters for anyone trying to judge whether footage is real or use these models for planning and robotics.

What does a typical video generator's training objective mainly reward?

Models are trained to predict or denoise frames so the output looks like the training footage. Physical correctness is not measured directly.

Why are rare events like glass shattering especially hard for video models?

Events that are uncommon in the data and physically complex give the model fewer examples to learn from, so its output is less reliable.

The guide describes object permanence failures mainly as what kind of problem?

To keep a hidden object consistent, the model has to carry its identity through frames where it is invisible, and limited context makes that unreliable.

What is the main reason AI video clips are often only seconds long?

More frames mean more tokens, and attention cost grows fast with token count, so length is limited by compute.

Which statement reflects a misconception about video generators?

Video generators have no explicit physics simulator. Any physics they show is implicit in learned patterns.