Paerrin retweeted
pearl-2:native LATEST UPDATE
FP8 IS LIVE ON TESTNET2
The FP8 fork activated at block 150283 (764f8c01eef28c2df5ee31fe3942b9ed8ac5d5863801fe0bdd40b04310b45b93). Testnet2 is running pearld 1.6.0-fp8-testnet1.
Image: ghcr.io/pearl-research-labs/…
Digest: sha256:7704d71db70cee34d3774e521d63395e3ab833036b449082135658ba3580b89f
github.com/pearl-research-la…
We aim to activate FP8 on mainnet in 4 to 8 weeks. We'll use that time to test and validate the scheme with the community and within the team. We expect this to be the final fp8 protocol update before mainnet unless testing uncovers a major issue.
The testnet miner is a complete reference implementation of the FP8 protocol. We're continuing to optimize mining and inference performance. Expect miner performance updates both before and after mainnet activation.
FP8 implementation PR #311
github.com/pearl-research-la…
Updated PIP-3 draft (V4 certificates)
github.com/pearl-research-la…
FP8 whitepaper
pearlresearch.ai/research/wh…
Paerrin retweeted
Building a distributed sensing network takes patience and a lot of individual operators making a small commitment. We see you. Thank you.
Paerrin retweeted
big milestone for pearl-2:native
FP8 testnet activation height reached at block 67,370
the v1.6.0 FP8 build is ready, with V4 certificates replacing INT7 mining proofs with FP8 matrix operations
first valid FP8 blocks still pending confirmation
one step closer to mining during AI inference
Paerrin retweeted
This is what is called preconditioning you for a false flag attack. These assholes want war so bad they're gonna pull some bullshit to keep it going.
Paerrin retweeted
The Generalized Bulletproofs implementation for Full-Chain Membership Proofs (FCMP++) has successfully been audited by zkSecurity!
'It proves that the Generalzed Bulletproofs protocol, in the form Monero implements it, is complete, zero-knowledge, and sound.'
zkSecurity Reviews Generalized Bulletproofs and Publishes a Lean Proof of Its Security (@zksecurityXYZ)
magicgrants.org/2026/11/12/G…
Paerrin retweeted
4/ Our commitment to data integrity remains.
We'll continue investing in our proprietary technology to deliver accurate, reliable market data across blockchain ecosystems, while upholding transparent, publicly accessible methodologies.
Our website, app and API will continue operating as usual.
Read the full announcement 👇
gcko.io/9dl2r3w
Paerrin retweeted
SDR (software-defined radio) turns a general-purpose radio front-end into whatever receiver the software tells it to be - the same hardware can track GNSS, ADS-B, or ambient RF. #SDR
On November 12, Outlook, Teams and Copilot went down in 74 countries for more than 12 hours.
When everything runs on one cloud, everything fails together. Including the agents meant to keep the business running.
A network with no single operator has no single outage.
Read more: acurast.com/blog/network-and…
Paerrin retweeted
Bitcoin hit $62K over the weekend and filled its downside fair value gap. I'm looking at extreme oversold conditions, a new gap toward $74,217 and three of five core on-chain signals in the bottom zone. I see an accumulation opportunity, but that doesn't mean we're done falling or that a bounce has to happen immediately.
📈 Kalshi Perpetuals
Trade Kalshi Perpetuals: kalshi.com/p/mooninpapa
I’m an independent Kalshi affiliate and may earn a commission. Perpetuals trading carries a high risk of loss; availability and eligibility may vary.
ETH is deeply oversold too, while ETH/BTC, BTC.D, ETH.D, SOL.D, stablecoin dominance, OTHERS.D, TOTALE50.D, TOTALE100.D, TOTALES and OTHERS show how uneven this selloff is. I compare TOTAL3ESBTC, OTHERSBTC and OTHERS/ETH, then the Nikkei comparison, DXY, EUR/USD, USD/JPY, S&P futures, VIX, oil, gold, silver and PAXG/BTC. Extreme fear isn't a timing signal on its own; I'm watching RSI resets, volume and TBO confirmation.
I review XMR, NEAR, QNT, JUP, VVV, NIGHT, DRV, FF, PIEVERSE, BEAM, SOL and ASTER, then Club picks AIOZ, FET, LTC, ONDO, PEPE, QUBIC, STRK, VIRTUAL, ZEC and ZIG. Stocks and ETFs include AMZN, BN, CIBR, DRUP, E3B, EOAN, INTC, ENRG, GSK, LSX, MRVL, PAVE, S0LR, UMDV, V, IREN, CRCL, COIN, MSTR, MU, AMD, IONQ, RKLB, SNDK and BMNR, alongside ENA and XRP. I finish with RAIL, LIT, AMP, ADA, PUMP, IUIT, VTI and SPY. Some charts are flashing reversal clues; others remain strongly bearish, so I'm separating an oversold opportunity from a confirmed trend change.
CHAPTER MARKERS
00:00 Bitcoin's weekend selloff and new fair value gap
05:15 On-chain bottom signals and Ethereum
10:31 Stablecoin dominance and oversold market caps
17:01 Extreme fear, altcoin ratios and TradFi
22:02 Gold, silver, PAXG/BTC and Club resources
27:46 My picks: XMR, JUP, NIGHT and DRV
34:12 BEAM, SOL, ASTER and Club crypto picks
39:39 Stock and ETF watchlist
49:06 YouTube picks and final reversal signals
Paerrin retweeted
So let me get this straight.
2021: Biden's infrastructure law puts $5B into FAA air traffic control facilities. That means towers and buildings. It doesn't replace the decades-old copper-and-switches network that connects them.
2023: The FAA hires Verizon for $2.4B to build that replacement network. Two years later it still isn't running. Project 2025 calls for spinning air traffic control out of the FAA and privatizing it if possible.
2025: DOGE fires about 400 FAA staff, including maintenance mechanics at 18 air traffic control facilities that were already short on technicians. Musk pushes Starlink to replace Verizon. Newark's radar and radios keep going dark because of old copper wire and untested telecom links. Congress puts $12.5B toward the fix and calls it a down payment.
June 2026: Air Space Intelligence wins a 12-year FAA contract worth up to $875.9M to build SMART, an AI air traffic management system. Months earlier, ASI's PAC donated to the campaign of Duffy's son-in-law. The award was competitive. ASI's Air Force deal was sole-source.
November 10: SMART goes live at all 22 Air Route Traffic Control Centers, about 50 days after its first launch over Washington. The controllers' union says it wasn't involved in the rollout.
November 13: An update hits the private network that links every center. The network reboots and crashes, the old equipment chokes, and controllers get bad data. The result is the third nationwide ground stop in US history: more than four hours, nearly 8,000 cancellations and 10,000+ delays.
Officials blame old equipment. ASI and Duffy say SMART had nothing to do with it. DHS says it can't rule anything out. Nobody has said what the update was or who approved it. And an earlier review had already warned that this network was no longer supported and that one failure could cut off an entire region.
Anyone who has shipped software into legacy infrastructure knows the appropriate playbook:
⭐️ Map every dependency.
⭐️ Test against real failure modes.
⭐️ Run old and new in parallel.
⭐️ Roll out in stages.
⭐️ Change one major thing at a time.
⭐️ Keep a rollback you've actually tested.
⭐️ Keep the operators in the room.
I suspect a lot of newer engineers raised on vibecoding have never learned this. Prompting gets something new on screen fast, but it won't teach you how a decades-old network reacts to an update at 4pm on a Friday.
Speed is not the metric. In aviation, reliability and accountability have to be built in from day one. Bring back the adults in the room.
Paerrin retweeted
Pearl Research Labs has now officially announced PearlQuant, its KV cache compression method. It is @prlnet’s first post since Oct 21, and CEO Omri Weinstein calls it “the most efficient KV cache quantization on earth.”
The results match the write-up this account covered on Oct 23. On GLM-5.3 it stores 3.2 bits per value against 9.1 for FP8, about 65% less, with no statistically significant quality difference on four benchmarks. At 128 concurrent sessions on one 8×B200 node, p90 time to first token is 7 s against FP8’s 144 s.
Peak throughput is 24% above FP8’s. At light load, FP8 is about 8% faster, up from about 5%.
The announcement is about serving and doesn’t mention mining. Mainnet mining still runs on 7-bit integer math, with the FP8 upgrade still to come.
Each inference session has its own KV cache, so a smaller cache means more concurrent users per GPU.
Introducing PearlQuant, our new KV cache quantization technique: 65% less storage than FP8, better throughput and faster serving, same quality.
GLM-5.3 on 8×B200, 128 concurrent sessions:
⚡ p90 TTFT: 144s → 7s
📈 24% higher peak throughput
🎯 Quality on par with FP8
pearlresearch.ai/research/bl…


























