Words and phrases I want to never see on my timeline again
- Trenches
- Trench warriors
- Cook
- Cooking
- Cooked
- Daily Runner
- Coded
- Murad
- Golden Bull Run
- Memecoin Supercycle
- NFT mania
- Skill issue
- SocialFi
- Tokenized Culture
- Vape
- Hot Air Rises
- Just keep clicking
- Pico Top
- Pico Bottom
- Quant
- No crying in the Casino
- Mass adoption
- Terminal
- Don’t fade
- Don’t sleep
- Cope
- Seethe
- This is [insert person don’t care about] followed by thread
- Bark
- Runes
- Believe in something
- SBR
- SEC
- XRP
- Presale
- Buy the dip
- Builder’s Market
- LIBRA
- Left Translated Cycle
- Right Translated cycle
- Bro
- Bruv
- Bruh
- KOL
- Creator
- Wycoff
- That stupid chart with all the phases of the cycle
- Sell The News
- Doginal
- Celebrity Coin
- Meta
-Sphere
- Ketamine
- Yap
- Quack
- Alt Season
- BTC Dominance
Other than those things, you’re free to talk about whatever you want
Why should your AI work make money for someone else?
With Perspective AI, every time your agents are used, you get paid.
It’s your effort and your reward.
JUST IN:
We’re proud to announce Zama's next genesis operator: @luganodes.
Luganodes is a world-class, non-custodial, institutional staking provider with $3bn+ in assets, supporting 40+ PoS networks and maintaining 99.9% uptime. Known for exceptional reliability, Luganodes ensures secure and compliant staking solutions.
Learn more about the Zama Protocol and its operators in our litepaper: #operations-and-governance" target="_blank" rel="nofollow noopener">docs.zama.org/protocol/zama-…
🔔 See you tomorrow for the next genesis operator reveal.
BioFi – when biological data becomes the human voice
In the world of Web3, assets are not just money or NFTs.
The biggest asset – is human data: DNA, behavior, voice, emotions, and even how we interact with AI.
@BioProtocol understands that through projects like $BIOMEAI and $AUBRAI, this platform is opening new doors.
-Each individual can manage-encode-and earn value from their own biological data.
This is a huge step forward:
From Web2, where data is collected,
→ to Web3, where data is returned to its creators.
The combination of AI + Blockchain + Digital Biology not only creates new technology, but also reshapes humanity's digital freedom.
#BioProtocol#BioFi#AI#Web3Revolution#DataOwnership#BlockchainForGood#BIOMEAI#AUBRAI#FutureOfAI
India pays homage to Sardar Vallabhbhai Patel on his 150th Jayanti. He was the driving force behind India’s integration, thus shaping our nation’s destiny in its formative years. His unwavering commitment to national integrity, good governance and public service continues to inspire generations. We also reaffirm our collective resolve to uphold his vision of a united, strong and self-reliant India.
बीते 11 वर्षों में भारत के Maritime Sector में ऐतिहासिक परिवर्तन आया है। हमें गर्व है कि आज हमारे पोर्ट्स Developing World के सबसे Efficient Ports में गिने जाते हैं।
1/ Bio Ecosystem Science Updates
From clinical studies to real-world pilots, Bio's pipeline of research projects is rapidly advancing.
Track scientific progress across decentralized biotechs in the Bio Network 🧵↓
This changes the game:
• Galectin‑3 AD PoC Q1 Q3 2026 (ELISA/IHC)
• CNS‑ready inhibitors next
• NEURON‑GALE + IP aim at $30B+ market; @BioProtocol Ignition Sales
oml 1.0 is the latest update from @SentientAGI
it's not just a concept anymore, but a real scalable system for protecting open models
in the new paper scalable fingerprinting for llms, the team demonstrated how to embed tens of thousands of cryptographic fingerprints into a model with zero performance loss
no one had achieved that before
the problem with open source models has always been the same
once weights are public, anyone can copy, fine tune, or rehost them and claim ownership
older fingerprinting methods couldn't fix this
they capped at around a hundred fingerprints before collisions, distortion, or key leaks started happening
models degraded, and protection failed
oml 1.0 changes that completely
in this new version the team embedded 24 576 unique fingerprints into llama 3.1 8b
while fully preserving performance
no regression on standard evaluation tasks
the model behaves completely normal unless a secret key is activated
this means open models can now carry thousands of independent cryptographic signatures that can't be removed
how it works
oml 1.0 introduces perinucleus sampling
a new method for generating keys and responses from low probability yet human plausible tokens
the fingerprints blend naturally into the model's token distribution and don't affect its behavior
keys don't overlap or interfere
so thousands of them can coexist without collisions or leaks
to make fingerprints resilient, oml 1.0 uses several specialized training techniques
▹ targeted sft - fine grained training at low learning rates to bind keys without altering base behavior
▹ model mixing - weighted averaging with the base model to avoid catastrophic forgetting
▹ benign data mixing - combine regular data with fingerprint data to stabilize the distribution
▹ parameter expansion - extend intermediate layers with new parameters that store fingerprints
▹ distribution alignment - maintain instruction following quality and natural outputs
as a result, fingerprints survive fine tuning, distillation, and model merging
they're also resistant to collusion attacks or key hunting attempts
in short oml 1.0 proves that fingerprinting isn't just an experiment anymore
it's a working method for large scale model attribution
this unlocks new possibilities
▹ tracking provenance of models
▹ licensing open weights
▹ enabling secure collaboration for research and commercial use
the oml 1.0 research team includes pramod viswanath, himanshu tyagi, sewoong oh, andrew miller, lucas irwin and others
they're pushing open source ai toward being not only open but also secure and verifiable
next updates for oml will include
higher key capacity
stronger collision resistance
and faster, more reliable verification
oml 1.0 marks a shift toward an era where open models don't lose authorship or become vulnerable
they can finally be open and protected at the same time