CURRENT SCALING APPROACHES FOR LLMS ARE ENCOUNTERING SIGNIFICANT LIMITATIONS

Engineers are looking at photonics and neuromorphics for more efficiency

The present "brute force" use of serial hungry chips/systems means that the exploitative use of energy, water and land is unsustainable. Scaling is limited not just by architecture but by context window limits and memory bottlenecks, Quadratic attention costs and token findings is limited. The answer is to be much smarter and the technique of small language models and associated agents. The grand scheme of one all singing LLMS may hit its limits within two years. The photon, not the electron is a harder entity to control but its advantages are many. Hybrid light carriers and silicon may be a stop gap but it is only to use existing manufacture and not a long term solution.

---

SEE ALSO

• FRACTAL DYNAMICS

---

• ETHICAL AI AND ASI (IT'S OVER)

---

• THE TYPE AND USE OF MODERN ROBOTS

---

• NO JOBS, PURPOSE AND ALIGNMENT

---

• PATTERNS MAKE A UNIVERSE

---

---