BTC $68,403.62 +5.55%
ETH $2,081.09 +8.58%
BNB $617.35 +2.39%
XRP $1.07 +6.68%
SOL $81.68 +5.94%
TRX $0.3333 +0.46%
DOGE $0.0724 +3.19%
ADA $0.1795 +2.30%
BCH $205.15 +0.56%
LINK $9.88 +4.28%
HYPE $61.35 +3.29%
AAVE $92.58 +3.59%
SUI $0.6860 +4.66%
XLM $0.1645 +6.02%
ZEC $546.91 +7.01%
BTC $68,403.62 +5.55%
ETH $2,081.09 +8.58%
BNB $617.35 +2.39%
XRP $1.07 +6.68%
SOL $81.68 +5.94%
TRX $0.3333 +0.46%
DOGE $0.0724 +3.19%
ADA $0.1795 +2.30%
BCH $205.15 +0.56%
LINK $9.88 +4.28%
HYPE $61.35 +3.29%
AAVE $92.58 +3.59%
SUI $0.6860 +4.66%
XLM $0.1645 +6.02%
ZEC $546.91 +7.01%

seed

All
Article
Flash

first_img ByteDance discusses training a model with over 50 trillion parameters, the Seed model team adjusts the architecture

According to LatePost, ByteDance is discussing a large model with training parameters exceeding 50 trillion, surpassing Alibaba's Qwen 3.8-Max (24 trillion) and Moonlight K3 (28 trillion), making it the largest known plan in the country so far. This plan is still in its early stages and does not guarantee a final release. The new model is intended to be led by Xiang Liang, head of Seed Foundation, in collaboration with Shen Ke, who is responsible for the pre-training data of large language models. Seed is reorganizing, dividing responsibilities, and allocating resources based on this.Two weeks ago, ByteDance founder Zhang Yiming held a company-wide meeting with Seed head Wu Yonghui. Zhang reassured the team that training large models is inherently difficult and that it is acceptable to lag behind for a period of time, hoping to aim for the upper limits of intelligence and join the world's top tier. He acknowledged that programming is a key direction at present, advocating for the integration of Volcano Engine, Feishu, and Doubao resources to build computational power and data advantages, while reminding not to be led by a single hot topic. He praised Seedance's differentiated leadership and clearly opposed distillation, believing it is difficult to truly surpass and that AGI barriers should be built from a more fundamental level, stating that the company will continue to increase investment in AI.In the past six months, Seed's multimodal performance has been outstanding, with Seedance 2.0, Seedream, and others driving Volcano Engine MaaS, but the market response to the language model Seed 2.0 has been limited, and its lagging coding capabilities have affected the revenue structure. ByteDance has hired Guo Daye at a high salary to specialize in coding and has consolidated related resources. In the face of the industry's general trend of increasing model sizes, ByteDance hopes to achieve a leapfrog advantage with a larger scale while promoting the elimination of horse racing and breaking down departmental walls to concentrate efforts on tackling challenges.

153 stolen addresses contain 132.95 BTC, and researchers are still unable to reproduce the Coldcard attacker's seed

According to monitoring by Bitcoin News, new research published by @PraveenPerera shows that Coldcard attackers seem to first identify addresses with vulnerabilities, then sort them by the amount of Bitcoin held, starting to transfer in batches from the addresses with the highest holdings. The transfer software used was relatively crude.One address had 225 spendable UTXOs, and the attackers extracted exactly the latest 200, leaving the earliest 25, which included a UTXO worth 0.16 BTC. This aligns perfectly with the limitation of a blockchain API investigated by researchers, which defaults to returning 200 records, indicating that the attackers may have failed to load the next page of data. The software even spent a UTXO of 294 satoshis, reportedly increasing the transaction fee by about 2040 satoshis, with the spent amount significantly higher than the value of the UTXO itself.The authors of the study believe that the builders of this tool may have a better understanding of the account balance system than of the Bitcoin UTXO model. Although the attackers seem to have obtained the complete seed of the victims, at least 75 BTC still remain in other addresses derived from the same seed. The biggest suspicion currently is that among the 153 stolen addresses, there are still 132.95 BTC, and researchers have been unable to reproduce the seed behind these addresses, so it cannot be ruled out that the attackers obtained undisclosed private device data or candidate data.

Attestable completes AI verifiable technology breakthrough, secures $20 million in seed round financing

AI security startup Attestable has officially announced its establishment and completed a $20 million seed round financing. This round of financing was led by Jamin Ball of Altimeter Capital and Yonatan Mandelbaum of TLV Partners, with participation from institutions such as Halcyon Futures, Cerca Partners, and several investors.Attestable founder Yogi stated that as AI gradually enters critical infrastructure, national security, and large enterprise systems, how to verify the credibility of AI operations has become an important issue faced globally. The company aims to build a universal verification layer for cutting-edge AI laboratories, critical infrastructure, and national-level applications.According to reports, Attestable uses Zero-Knowledge Proof technology to shift trust in AI systems from data centers to mathematical verification mechanisms. This technology can prove that a certain approved model, model weights, input data, and operational strategies indeed generated specific outputs, while not disclosing model parameters or user privacy data, and without needing to rerun the model for verification.Attestable stated that its technology has currently achieved verification inference of the Meta Muse Glimmer 30B model on a single NVIDIA H100 GPU, reaching a speed of 85 tokens per second. The generated proof documents are small and possess quantum-resistant characteristics, allowing for rapid verification.In response, Ethereum co-founder Vitalik Buterin commented, stating that this achievement means the performance loss of Zero-Knowledge Proofs for large language models is approaching single-digit levels. He noted that the next challenge is to further reduce the performance overhead of technologies such as Fully Homomorphic Encryption (FHE).
app_icon
ChainCatcher Building the Web3 world with innovations.