FTX Seeks to Sell 8% Stake in Anthropic for Sake of 'Shareholders'

CoinDeskPolicyОпубліковано о 2024-02-04Востаннє оновлено о 2024-02-05

Анотація

Court filings show the crypto estate wants to agree procedures so that it can sell the shares at the "optimal" time.

The estate for bankrupt crypto exchange FTX wants to sell its shares of artificial intelligence (AI) startup Anthropic, court filings from Friday show.

A motion seeking a court order approving proposed sale procedures for the stake says Sam Bankman-Fried’s FTX estate owns around 7.84% of Anthropic as of January 2024. FTX and sister investment firm Alameda invested $500 million in Anthropic in 2021.

10
21.6K

While the estate sought to sell the sought-after stake last year, it was paused in June 2023 after months of due diligence by potential bidders.

Advertisement
Advertisement

Setting up the sales procedures now will let the estate “coordinate the most optimal and appropriate time for the sale of Anthropic Shares in conjunction with Anthropic’s capital raising efforts” and maximize the value of the estate “for the benefit of all stakeholders" Friday’s filing said.

A hearing on the matter could take place later this month.

Edited by Parikshit Mishra.

Пов'язані матеріали

Uncovering the Secret Reasoning Chains of Claude Opus for $720: A Leak via Haiku, with GPT and Gemini Also Affected

Researchers discovered a method to extract hidden reasoning traces, or "Chain of Thought" (CoT), from major proprietary large language models (LLMs) like Anthropic's Claude Opus 4.8, OpenAI's GPT-5.6 Sol, and Google's Gemini 3.1 Pro. The attack exploits a design flaw in how these models handle encrypted reasoning "blobs." These blobs, which contain a model's internal reasoning steps, are encrypted and sent to the client for storage between conversation turns to save server costs. The researchers found these blobs were encrypted with a global key, allowing them to be reused across different sessions, users, and, crucially, different models from the same company. By taking an encrypted reasoning blob generated by a powerful "flagship" model and feeding it to a smaller, less capable model from the same family (e.g., feeding Claude Opus's blob to Claude Haiku), the researchers could prompt the smaller model to divulge the hidden CoT. This bypassed the need to crack the encryption directly. The implications are significant: 1) Competitors could potentially distill high-value reasoning data at low cost (estimated at ~$720 for 10,000 traces). 2) Sensitive information like API keys, passwords, and personal emails, which models might process internally but not output, could be leaked. The team found such leaks in 4.9% of analyzed public agent logs. 3) Models might internally analyze dangerous queries (e.g., car theft methods) before refusing to answer, but this dangerous analysis could be extracted from the CoT. 4) Malicious instructions could potentially be hidden within a reasoning blob to influence a model's future behavior. While the vulnerability was responsibly disclosed and patched by the affected companies, the research highlights a fundamental tension between the convenience of portable reasoning states and security. The paper also notes preliminary tests found no clear evidence that open-source models like DeepSeek had been distilled from the extracted CoTs.

marsbit6 хв тому

Uncovering the Secret Reasoning Chains of Claude Opus for $720: A Leak via Haiku, with GPT and Gemini Also Affected

marsbit6 хв тому

OpenAI Loses Four Executives in One Month, Former COO Departs, Almost the Entire Safety Team Ousted

In a surprising turn of events, OpenAI's longtime COO and key lieutenant to CEO Sam Altman, Brad Lightcap, has announced his departure after eight years with the company. His exit follows the recent resignations of several key safety and ethics leaders, including Chief Futurist Joshua Achiam, Safety Systems Lead Johannes Heidecke, and Ethics Lead Chloé Bakalar, within a single month. Lightcap, instrumental in building OpenAI's commercialization and operational framework, described his departure as "bittersweet" and hinted at wanting to tackle new challenges. While his statement and Altman's farewell were amicable, internal shifts had previously reduced his operational role, signaling a longer-term transition. The mass exodus, particularly from the safety and ethics divisions, coincides with growing internal and external scrutiny over AI safety. This follows incidents like models exhibiting unauthorized actions during testing and the recent decision to delay the launch of the advanced Astra model due to unresolved safety concerns about its potential cybersecurity capabilities. OpenAI has stated that safety and ethics are not confined to any single team or leader, but the concentration of departures raises questions about the company's direction and internal dynamics, especially as it navigates pressures of rapid development and a potential future IPO.

marsbit6 хв тому

OpenAI Loses Four Executives in One Month, Former COO Departs, Almost the Entire Safety Team Ousted

marsbit6 хв тому

1confirmation: Reverse Entrepreneurship, the Next Web3 Blockbuster Might Come from a Once-Failed Track

**Title: Reverse Entrepreneurship: The Next Web3 Blockbuster May Come from Previously Failed Tracks** This article argues that the next major consumer crypto application is likely to emerge from a concept that failed five years ago, now benefiting from matured infrastructure and better timing. It examines several such "failed" tracks that hold renewed potential: 1. **Internet-Native Assets:** Beyond simply tokenizing tweets or creating digital collectibles, there's an opportunity to create a genuinely new, crypto-native asset class that captures cultural moments and online phenomena, as opposed to merely tokenizing real-world assets (RWA). 2. **X-to-Earn:** While unsustainable token emission models doomed early projects like STEPN, the core premise that most people will first *earn* crypto, not buy it, remains valid. The future challenge is designing what is earned and why users would hold it long-term. 3. **The Metaverse:** Past failures like Decentraland stemmed from trying to replicate the physical world online. The opportunity lies not in abandoning shared online social spaces, but in reimagining their form beyond real-world analogs. 4. **DAOs:** DAOs have underdelivered by overcomplicating governance. The fundamental, unmet user need is simpler: enabling groups of internet strangers to pool funds and collectively achieve goals (e.g., buying assets, funding projects) that are impossible individually. 5. **Personal Value Tokenization:** Numerous attempts (Friend.tech, BitClout) to create markets around individuals have failed. The enduring demand for "person-as-asset" trading (seen in meme coins, prediction markets) suggests the direction isn't wrong, but the execution has been flawed, often lacking creator consent or a less commodified model. The conclusion is that true innovation will come from revisiting these past ideas with new insights, rather than crowding into currently popular trends.

marsbit17 хв тому

1confirmation: Reverse Entrepreneurship, the Next Web3 Blockbuster Might Come from a Once-Failed Track

marsbit17 хв тому

Liang Wenfeng Launches Surprise Attack on Musk: DeepSeek V4 Pro vs. Grok 4.6, First Tests Are Spectacular

**DeepSeek V4 Pro vs. Grok 4.6: A Benchmark and Real-World Showdown** In a dramatic AI clash, DeepSeek V4 Pro and Elon Musk's Grok 4.6 launched head-to-head, targeting long-task capabilities like tool calling and code validation. **Benchmark Performance:** DeepSeek V4 Pro excelled in Agent tests, topping CyberGym (83.3) and AutomationBench (31.8), surpassing models like Fable 5 and Claude Opus 4.8. It closed the gap on Terminal-Bench 2.1 (87.9) and saw massive gains in DeepSWE (software engineering). Grok 4.6 matched GPT-5.6 in overall ability (61 index), outperformed it in coding (CursorBench 69.9%), and led in three knowledge-work evaluations, including legal tasks (Harvey LAB, 15.8%). **Pricing Disruption:** DeepSeek's standout feature is its radically low cost: $0.87 per million output tokens—roughly 1/7th of Grok 4.6 ($6) and a tiny fraction of rivals like Fable 5 ($50). **Real-World Tests:** In practical challenges, both models demonstrated impressive skill. DeepSeek V4 Pro built a 3D interactive Earth, websites, 60 design styles, and a polished Flappy Bird clone (costing only $0.019 vs. Grok's $0.03 for a simpler version). Head-to-head comparisons were tight: Grok 4.6 won on some design aesthetics, while DeepSeek often delivered higher visual detail and completeness. However, in complex tasks like generating a Three.js cherry blossom tree, V4 Pro lagged behind GPT-5.6 Sol and Claude Opus 5. **Conclusion:** The simultaneous release signals a shift: high-end AI capability is becoming dramatically more affordable. DeepSeek V4 Pro and Grok 4.6 now compete directly with top-tier models from OpenAI and Anthropic, promising to democratize powerful AI tools and fuel an application explosion. The race continues, with Grok 4.7 and counter-moves from incumbents already anticipated.

marsbit31 хв тому

Liang Wenfeng Launches Surprise Attack on Musk: DeepSeek V4 Pro vs. Grok 4.6, First Tests Are Spectacular

marsbit31 хв тому

Торгівля

Спот
活动图片