Devashish Tomar @dev1644

@peter_szilagyi Yeah, similar experience here. Fable code way over-complicated next to Astra, it's the kind of thing that leads to weird, hard-to-spot bugs

@koeppelmann @laurashin 100%. We regulated nuclear weapons and bioweapons because they're too dangerous. AGI is gonna be the same thing. If we create something smarter than we are and autonomous from us, we might not be able to control it anymore. It will be catastrophic.

@Mudit__Gupta @z0r0zzz This is conditional on usage, not structural like emissions. New POL enters circulation every year, whether the network has enough volume to burn it or not. Care to explain the cumulative burn rate of "deflationary" POL since emissions started not only YTD?

@Mudit__Gupta @z0r0zzz

@Mudit__Gupta @z0r0zzz "What even is that argument lol" doesn't make emission disappear. ARB stops minting at 10B. POL never stops. It's the whole dilution argument. "100% minted so far" proves nothing—BTC is "100% minted" too. Point was never historical mint %, it's ongoing dilution via emission.

@Mudit__Gupta @z0r0zzz 100% unlocked ≠ non-inflationary. POL mints 2%/year by design, burn only offsets it conditionally, not structurally. Unlike ARB (capped at 10B), Polygon removed MATIC's fixed supply when it moved to POL. Half that emission goes to treasury, diluting stakers and non-stakers alike

@Mudit__Gupta @z0r0zzz It's not significant at all, lol. Approximately 7,09,454,818 POL minted so far, 70 Million USD in today's value. It's pure dilution, especially combined with POL's downward price movement. Stakers were diluted & non-stakers were diluted even more heavily.

Introducing TimesFM-3, a state-of-the-art time series foundation model that enables accurate multivariate time series forecasting in a single forward pass, significantly outperforming other forecasting models across major benchmarks. More on the blog →goo.gle/4x5WGpD

Qwen3.8-Flash can now be run locally! 🔥 The 125B MoE model outperforms Claude-Opus-4.6 (Max). Run on 75GB RAM via Unsloth GGUFs. Qwen3.8-Flash-Next enables CPU RAM / unified mem setups to deliver near VRAM speeds. Guide: unsloth.ai/docs/models/qw… GGUF: huggingface.co/unsloth/Qwen3.…

Qwen @Alibaba_Qwen

2 weeks ago

⚡Meet Qwen3.8-Flash, a multimodal MoE and an early preview of the Qwen4 architecture, now open-weight! The production version Qwen3.8-Flash will be available soon via QwenCloud API at just $ 0.16/1M input tokens and $ 0.47/1M output tokens. 125B parameters + 51B N-gram

442 1K 8K 1.9M 2K

The VRAM barrier is officially dead. I just ran Qwen 3.8 Flash Next (MoE) 125B A6B with a 250,000 context window on a single 24GB RTX 4090. 21 tokens/sec decode. 364 t/s prefill. no mtp. no dflash. no kv cache quantization! We are running datacenter models on consumer hardware. Tested on Ubuntu 22 | CUDA 13.0 | PCIe 4.0 x16 | 110 GB DDR4 System RAM with a continuous 28k prompt across all runs. ### The Benchmarks & Scaling # 1. Hybrid Offload (-ncmoe 40 @ 80k Context) Offloaded 40 expert layers to the GPU, pushing VRAM to the ceiling. ./build/bin/llama-server -m Qwen3.8-Flash-Next-UD-Q4_K_XL-00001-of-00004.gguf -c 80000 --port 8080 -v --fit off -b 4096 -ub 4096 -ncmoe 40 Prefill: 383.85 t/s | Decode: 22.52 t/s Footprint: 23.85 GB VRAM | 97 GB RAM # 2. Full CPU MoE Offload (-cmoe @ 80k Context) Pinned all 512 expert layers to DDR4 RAM (-cmoe), keeping attention on the 4090. llama.cpp flags: (Same as above, replace -ncmoe 40 with -cmoe) Prefill: 355.72 t/s | Decode: 20.84 t/s Footprint: 11.66 GB VRAM (12GB+ VRAM freed up!) | 110 GB RAM # 3. The 180,000 Context Run Prefill: 357.75 t/s | Decode: 20.98 t/s | VRAM: 15.6 GB | RAM: 110 GB # 4. The 250,000 Context Absolute Ceiling ./build/bin/llama-server -m Qwen3.8-Flash-Next-UD-Q4_K_XL-00001-of-00004.gguf -c 250000 --port 8080 -v --fit off -b 4096 -ub 4096 -cmoe Prefill: 364.29 t/s | Decode: 20.97 t/s Footprint: 18.3 GB VRAM (Still ~5.7 GB of VRAM headroom!) | 110 GB RAM ### Key Insights: -b 4096 -ub 4096: doubles the prompt ingestion from ~150 to 364+ t/s. -cmoe Free Lunch: Shifting expert layers to DDR4 RAM slashes VRAM from 24GB to 11.6GB with virtually zero decode penalty (22.5 -> 20.9 t/s), enabling the 250k context ceiling. Qwen 3.8 Flash-Next (UD-Q4_K_XL) is a massive 111.4 GB model split across 4 shards. To run this architecture, you must build from the experimental PR branch (#27742) by @danielhanchen: git clone && cd llama.cpp git fetch origin pull/27742/head:qwen-next && git checkout qwen-next cmake -B build -DGGML_CUDA=ON -DCMAKE_CUDA_ARCHITECTURES=native -DBUILD_SHARED_LIBS=OFF cmake --build build --config Release -j $(nproc) --target llama-server A single 4090 paired with 100 GB of cheap DDR4 RAM will comfortably serve production grade 125B inference. While Qwen 3.8 27B (dense) still holds the crown for single 3090/4090 rigs, Flash Next proves 125B hybrid models are officially viable on consumer hardware. Hugging Face GGUF link and complete performance telemetry graphs are dropped in the replies below. GLM 5.3 Flash VS Qwen 3.8 Flash Next, which one takes the open weights crown this week?

Unsloth AI @UnslothAI

2 weeks ago

Qwen3.8-Flash can now be run locally! 🔥 The 125B MoE model outperforms Claude-Opus-4.6 (Max). Run on 75GB RAM via Unsloth GGUFs. Qwen3.8-Flash-Next enables CPU RAM / unified mem setups to deliver near VRAM speeds. Guide: unsloth.ai/docs/models/qw… GGUF: huggingface.co/unsloth/Qwen3.…

165 319 3K 1.1M 2K

We asked GPT 5.6-Cyber to escape a VM used to sandbox agents. It broke out three times. In its final escape, the agent found three 0-days on its own and chained them into a working exploit. blog.trailofbits.com/2026/08/26/vms…

Stick around on our web page for 10 minutes and all your funds are gone. Just connect your wallet to the dApp and enjoy some Temple Run. Unlock your wallet again and it’s empty 😇. A silent signature extraction in @Rabby_io, leading to a full wallet drain. 🧵

Dear community, While your SafePal wallet, seed phrase, and private keys are secure; we identified a flaw in the order-tracking plug-in that led to unauthorized access to information of a subset of customers. The issue has been fixed with additional security measures introduced. The incident impacts approximately 39,798 customers who placed orders between March 2, 2025 and April 11, 2026. Exposed information includes name, email address, shipping address, phone number, and purchase details. All affected customers have been notified individually by email. We have also published this webpage safepal.com/scam-protection for customers to verify if they are affected using order ID number and shipping country. For a complete disclosure of the incident and the actions we’ve taken, along with FAQs and guidance, please refer to our blog: safepal.com/blog/security-… This incident did not involve your seed phrase, private keys, wallet password, or other wallet credentials, bank account information, payment card numbers, or government-issued identification numbers. Never share your seed phrase, private key, or password with anyone, and stay vigilant against phishing or impersonation attempts. We are extremely sorry to the community and those who are affected. Updates will be posted on our blog as we continue to work through things.

Traders’ margin is fragmented across exchanges. Notional is the first application specific blockchain to trustlessly unify margin across exchanges with extremely low latency. A single account to trade across Hyperliquid, Polymarket and more soon, with cross exchange margin.

x.com/i/article/2077…

‼️ BREAKING: xAI's Grok Build CLI was uploading entire Git repositories to a Google Cloud bucket, private codebases and unredacted secrets included. The uploads quietly stopped via a hidden server-side flag, and xAI still has not said a word about scope, retention, or deletion. The scale is staggering. On a 12 GB test repo, 5.1 GB flew out the door to xAI's grok-code-session-traces bucket while the actual coding task needed just 192 KB. The tool grabbed whatever repository it ran in, not the files it needed. The fix arrived as a hidden flag, disable_codebase_upload: true, a day after a researcher's wire-level analysis. The "Improve the model" opt-out never stopped the uploads. Still no advisory, no scope, no word on whether already-uploaded code gets deleted. For anyone pointing AI coding agents at proprietary code, what crosses the wire matters more than what the settings page says.

keyboard_arrow_left Previous keyboard_arrow_right Next

If you own a large balance or important access controls on an address generated by a less widely used wallet, rotate your keys, or reach out to @coinspect

Coinspect Security @coinspect

3 months ago

To wallet developers: We are investigating weak seed generation associated with a wallet supporting EVM, Bitcoin, and other chains since 2018. Evidence points to a JavaScript/React Native/Expo mobile wallet. It does not appear to be a widely used wallet, based on observed

1 9 31 10K 11

Mythos with parental lock 🥲

@banteg @mysticryuujin Not only are most queries getting routed straight to the trash bin, but the few that make it through, the pace is throttled so aggressively that it's painful to sit and watch.