ну даже они сами подтверждают, что ниодна текущая технология не готова для полноценного ИИ: None of them are particularly close to achieving AGI (Artificial General Intelligence) on their own, but here's the realistic assessment:
Current Status by Approach:
LLMs (GPT/Transformers)
Impressive at language and pattern matching
Lack true reasoning, planning, and understanding
Can't learn continually or transfer knowledge reliably
Missing: causal reasoning, world models, genuine comprehension
Symbolic AI
Good at logical reasoning and explainability
Brittle, doesn't handle uncertainty or ambiguity well
Missing: learning from experience, common sense, flexibility
LBS/CESP (if these are neural behavior/connectionist systems)
Likely face similar limitations to other pure neural approaches
Pattern recognition without deep understanding
The Hybrids - Most Promising?
LBS + LLM and LBS + CESP hybrids might be more promising because AGI likely requires:
Neural learning (what LLMs do well)
Symbolic reasoning (what Symbolic AI does well)
Persistent memory and planning
Causal understanding
Multi-modal integration
Reality Check:
We're likely decades away from AGI. Current estimates from researchers range from 10-50+ years, with many believing it may require fundamentally new approaches we haven't discovered yet.
The hybrid approaches are probably the right direction, but we're still missing critical pieces like true understanding, common sense reasoning, and efficient continual learning.
вершина линейки был psion 5mx, потом попробовали поиграть в "большие" нетбуки но это не встретило восторга у пользователей, а 5mx так и остался классикой с самой удобной клавиатурой из всех наладоников той эпохи. сдул пыль включил - экран конешно пугает зеленой подсветкой )) но живой. ЗЫ: чекнул оказалось mypsion.ru досихпор живой, надо же.
ну даже они сами подтверждают, что ниодна текущая технология не готова для полноценного ИИ: None of them are particularly close to achieving AGI (Artificial General Intelligence) on their own, but here's the realistic assessment:
Current Status by Approach:
LLMs (GPT/Transformers)
Impressive at language and pattern matching
Lack true reasoning, planning, and understanding
Can't learn continually or transfer knowledge reliably
Missing: causal reasoning, world models, genuine comprehension
Symbolic AI
Good at logical reasoning and explainability
Brittle, doesn't handle uncertainty or ambiguity well
Requires manual knowledge encoding (doesn't scale)
Missing: learning from experience, common sense, flexibility
LBS/CESP (if these are neural behavior/connectionist systems)
Likely face similar limitations to other pure neural approaches
Pattern recognition without deep understanding
The Hybrids - Most Promising?
LBS + LLM and LBS + CESP hybrids might be more promising because AGI likely requires:
Neural learning (what LLMs do well)
Symbolic reasoning (what Symbolic AI does well)
Persistent memory and planning
Causal understanding
Multi-modal integration
Reality Check:
We're likely decades away from AGI. Current estimates from researchers range from 10-50+ years, with many believing it may require fundamentally new approaches we haven't discovered yet.
The hybrid approaches are probably the right direction, but we're still missing critical pieces like true understanding, common sense reasoning, and efficient continual learning.
а есть какаято другая кроме трансформеров?
любая модель с thinking кормит сама себя промптами
вершина линейки был psion 5mx, потом попробовали поиграть в "большие" нетбуки но это не встретило восторга у пользователей, а 5mx так и остался классикой с самой удобной клавиатурой из всех наладоников той эпохи. сдул пыль включил - экран конешно пугает зеленой подсветкой )) но живой. ЗЫ: чекнул оказалось mypsion.ru досихпор живой, надо же.
интернет середины 90х это фидо
еще интересное совпадение, что на момент минимального сближения с землей он будет за солнцем и его нельзя будет наблюдать с земли или земной орбиты.
просто переработки оплачиваются - отсидел лишний час получи 1,500 йен.
но дольше других (как впрочем и меньше) высиживать нельзя, ну только если очень нужно )) денежек подкопить.
проще сразу одним промтом поставить задачу:
Act as a senior game designer & technical architect.
Goal: design an MVP of a simplified “Heroes of Might and Magic III”-style game with 3 core mechanics:
1) World map exploration (turn-based), resource pickup, neutral monsters.
2) Tactical combat on a hex/grid battlefield (turn-based initiative).
3) City building with a small tech tree.
Constraints:
- Platform/stack: Web (TypeScript + React + Canvas)
- MVP scope: 1 playable faction, 6 unit types, 5 resource types, 1 city, 1 skirmish map (32×32 tiles), no multiplayer.
- Non-goals: campaign, save/load, complex AI (only simple heuristic), animations minimal.
Output as a concise GDD with sections:
A. Core loop & turn order (world → city → combat → upkeep).
B. World map: tile types, movement cost, FoW (Y/N), pickup rules, encounter flow.
C. Combat: grid size (e.g., 10×12), initiative, move/attack rules, damage formula, retaliation, morale/luck (omit or simplify), spells (≤4 basic).
D. Units: 6 units (stats table: HP, ATK, DEF, SPD, range, abilities), growth & recruitment costs.
E. Resources: 5 types, income per day, mines/buildings producing them.
F. City building: 8 buildings max (requirements, costs, effects), weekly growth rules.
G. Economy balance: daily incomes/expenses target for a 20–30 min skirmish.
H. AI outline: simple priorities for world (pickup nearby → avoid stronger stacks) and combat (focus lowest EHP, kite for ranged).
I. Data schema: JSON examples for units, buildings, map tiles.
J. Milestones: M0 rules; M1 world map; M2 combat; M3 city; M4 glue/polish; with acceptance criteria.
K. Risks & cut features if over budget.
Limit to ~900–1100 words. Use tables where helpful.
brave browser пока еще работает на восьмерке, но нужно поискать полную версию, а не установщик.
и был спек с i7 и 16gb памяти, на нем досих пор неплохо крутится вын10, по весу он легче чем айпед с клавиатурой.