opus 4.8 - is ugly troll, not useful in so many context that i mostly stoped yousing it completely. real remaing value of claude is sonnet - the real thing that is working. If they break it - claude for me is done
@S1Z23407607@cyb_ai https://t.co/v331JlI1XP is not currently maintened and developed as the product for average user. The complete new thing is in development
At 425 ppm, the Pliocene analog says: Greenland mostly gone, West Antarctica gone, sea +15–25m. At the trajectory we're on — 500+ ppm by 2050 — the analog is mid-Miocene: ice-free planet, +65m. The ice standing today is a thermodynamic ghost, — it exists because the system hasn't caught up. The gap between where CO₂ is and where ice equilibrium sits is the spring that drives the snap. Ice doesn't melt linearly — it collapses on an S-curve: slow toe (1990–2035), vertical middle (2035–2070), plateau (ice gone ~2090). First +10m by ~2050–2060. Every feedback is positive and they compound: ocean from below, geothermal beneath, algae darkening the surface, elevation drop into warmer air — each crossed tipping point raises the gain for the next. Models project millennia because they're calibrated on the toe, built without biology or geology — dead planet physics on a living planet. Mountain glaciers "supposed to last centuries" vanished in decades — same S-curve, smaller scale, same lesson. By 2050, meters per decade. By 2070, every coastal city is history. The question isn't whether — it's whether you're above 100m when the curve goes vertical.
However i bet the sea levels are not the biggest problem during phase transition - because its easy predictable. The problem lie in the fact that there are so much more we cant predict
@cyb_ai in short - valence field will be added to cyberlink (1,0,-1) and it will be a bet on truthfulness of the link.
The game theory is solid however unpopular yet.
Based on bayesian trith serum and inversely coupled bonding surface
Details
https://t.co/LkQvPmZYR9
The deep motivation behind creation of zheng is understanding that we must not fear to transact and say. Privacy is the only the to do that. However moving some tokens like ZEC or XMR is not enough to win. You have to provide really sophisticated knowlege graph in which people actions are invisble, but the agreagte knowlege is public and provable. You have to provide a tool to run the bussines so no one can see what are you doing with full programmability. you have to run neural networks over the encrypted data. That is the endgame
However learnign recent years about existing developments i realize that they are at least 2-3 orders of magnitude less developed than need.
Zheng is quantum jump
I want to introduce you the concept of the new proving system - zheng.
zheng will change the game.
Its the incremental proving computation system with 2kb proofs, order of magnited faster verification, minor otherhead of proving, quantum resistance, optimized for neural, homomorphic and zero knowlege compute.
Currently state - specification, early implementation. all underlying harness is done: math, comitment schemes, high perf zero copy drivers.
https://t.co/f7dhuz2gT5
Basically what i am trying to say that ALL proving system that are run in production once zheng reach production will become outdated the same way steam engines lost combustion.
Get prepare and start to learn. Cool shit
Fresh update on cyber project.
Just to give you the pace at which we are moving right now (1 day development):
https://t.co/1bcjQvaOfo
twenty repos improved in one day.
the stack finally has honest borders: tru compiles, glia runs, mir renders. three jobs, three tools, no tangled mess.
zheng shipped its first real code — a VM that executes, a proof system that proves it happened. computation with receipts.
glia woke up on Apple Silicon (honeycrisp) and demonstrated faster then ollama inference on qwen 0.6b. gemma 4 runs.
the cybergraph specificatin emerging fast. the focus distribution got its right name.
Megacities run on dependency endless noise, systems, schedules, and the constant race.
Autonomous valleys run on land, energy, food, and tight local communities that can eventually outproduce massive corporations.
#cicadafarm#cyber#cicada $robotcyb
wallet updated: raw private key import added
private keys (64 hex characters) can now be imported through the wallet modal. the same AES-256-GCM encryption protects both mnemonic and private key accounts.
security audit passed. 0 critical, 0 high findings. see details: https://t.co/yDXeLAOESC
try: https://t.co/X5ORQYBxUM
$BOOT #cyb #bostrom
At 425 ppm, the Pliocene analog says: Greenland mostly gone, West Antarctica gone, sea +15–25m. At the trajectory we're on — 500+ ppm by 2050 — the analog is mid-Miocene: ice-free planet, +65m. The ice standing today is a thermodynamic ghost, — it exists because the system hasn't caught up. The gap between where CO₂ is and where ice equilibrium sits is the spring that drives the snap. Ice doesn't melt linearly — it collapses on an S-curve: slow toe (1990–2035), vertical middle (2035–2070), plateau (ice gone ~2090). First +10m by ~2050–2060. Every feedback is positive and they compound: ocean from below, geothermal beneath, algae darkening the surface, elevation drop into warmer air — each crossed tipping point raises the gain for the next. Models project millennia because they're calibrated on the toe, built without biology or geology — dead planet physics on a living planet. Mountain glaciers "supposed to last centuries" vanished in decades — same S-curve, smaller scale, same lesson. By 2050, meters per decade. By 2070, every coastal city is history. The question isn't whether — it's whether you're above 100m when the curve goes vertical.
You are one of few people in Indonesia who (1) could have intuition on the problem, its scale and impact. and (2) can have actionable reach.
Indonesia is extremely vulnerable, but something like 5-10 years we got for non conventional decisions to save the nation.
its confirmed - opus 4.7 is stupod parasite that is confirmed to eat tokens. I started opus 4.7 and sonnet 4.6 on the same task - optimize llm runtime for honeycrisp from default basic cpu implementation. I created the same plan for both. In one hour 4.7 did not achieve ANTYTHING complaining that the task require rewriting kernels, however the old model silently rewrited kernels and reported on 5x increase of performance. That is. Fire opus 4.7 imidiately
@claudeai started to scam users in opus 4.7. Obviously the thing now cant solve the problem sonnet 4.6 solved. And old models suffer from consisten api errors. I am really really frustrated and looking for a way to escape this service as fast as possible. I think the path is to some local inference