Revealed: the Alameda venture capital portfolio

Financial Times發佈於 2022-12-07更新於 2022-12-08

文章摘要

A look inside Sam and Caroline’s grab-bag of crypto, etc.

As well as running a crypto exchange that didn’t exchange crypto and owning a hedge fund that didn’t hedge, Sam Bankman-Fried had a venture capital fund that didn’t venture its own capital.

The VC division, in contrast to the rest of the FTX group, can now provide some insight into where some of the money went. Here’s where it went:

The screenshot above and all those below are taken from an Excel spreadsheet dated early November, when SBF was seeking rescue funding amid a run on FTX customer deposits. Mobile and app users are advised to click the magnifying glass below each image for more detail.

According to a person familiar with the rescue effort, the document shows an Alameda Research private equity portfolio — with some FTX bets mixed in — that was being offered as collateral in an attempt to secure a new credit line for the stricken group.

The disparate bundle of nearly 500 illiquid investments is split across 10 holding companies. The total investment value is given on the spreadsheet as in excess of $5.4bn.

As well as forming a central plank in efforts to maximise recoveries from FTX’s bankruptcy, the portfolio might offer regulators insight into whether the group’s trading and exchange businesses were ever operationally separate as claimed.

Bankman-Fried conceded in an interview with the Financial Times that he was involved in Alameda’s venture capital activities but has so far ducked questions around the misuse of FTX customer funds.

Going by the spreadsheet, boundaries between SBF’s companies were blurred. Two of Alameda’s biggest holdings, the crypto miner Genesis and the artificial intelligence research group Anthropic, are also listed on the draft FTX balance sheet published last month by FTAV. (Semafor subseqently reported that FTX had seized certain assets from Alameda after a margin call.)

As previously reported, the portfolio includes stakes in FTX backers Sequoia Capital and Anthony Scaramucci’s SkyBridge Capital, as well as in Elon Musk’s SpaceX and Boring Company projects through the investment in K5.

Of Alameda’s remaining investments, crypto and DeFi projects account for the majority. But the list also includes numerous start-up video game studios and betting platforms, online banks, publishers, a fertility clinic, a military drone maker and a vertical farming company,

Some entries have no clear link to an active business, suggesting they may be misspelt or mislabeled.

Note that FT Alphaville has excluded entries where an investment type is not given, which removes approximately a dozen names with a total stated investment of about $100mn. All other data are presented as they were shown to prospective FTX investors. The FT makes no claim as to the data’s accuracy or completeness.

When asked about why FTX used customer funds to prop up Alameda, SBF has repeatedly pleaded ignorance. The former FTX CEO said that to avoid conflicts of interest he chose not to get involved in Alameda’s trading and risk management, so before last month was not fully aware of its parlous state.

However, SBF told the FT that in early summer he had participated in conversations where Alameda’s financial health and borrowing were discussed. The venture capital investments Alameda had made were “effectively, some of them, on margin”, he added.

Alameda’s spreadsheet predates SBF’s current media blitz by a month, though it comes with all the same warnings about potentially selective recall and unreliable presentation, As the FT’s Joshua Oliver reports:

Bankman-Fried’s attempt to account for what went wrong was laced with caveats and references to his incomplete memory. He cited lack of “confidence” in his answers at least a dozen times, calling other responses “idle speculation” or “shitty answers”. At one point, he paused for half a minute with his head in his hands.

Caroline Ellison, former CEO of Alameda and SBF’s one-time romantic partner, could not be reached for comment.

你可能也喜歡

三年之后:回看 2023 年我对 ChatGPT 的判断

2023年3月,在GPT-4发布前,作者王健硕对ChatGPT的未来做出了二十项预测。三年后的2026年,他使用AI agent对这些判断进行回顾验证,评估其准确性。 **主要看对的方面:** 1. **RAG成为主流**:预测通过外部检索而非修改模型来解决知识更新和幻觉问题,这已成为行业标准架构。 2. **LUI(自然语言界面)兴起**:预测自然语言交互将催生巨大新产业,Agent、MCP协议等发展印证了这一点,但LUI是与GUI共存而非取代。 3. **机器人网络与新寻址系统**:预测Agent将自动协作并需要新寻址方式,MCP、A2A等协议正朝此方向发展。 4. **中国能做出可用大模型**:预测中美模型差距会迅速缩小,事实如此,但真正领先的国内厂商与其当初点名不同。 5. **AI无意识与图灵测试本质**:核心判断“AI无意识,图灵测试仅测表象”基本成立,但“绝无意识”的绝对论断被后续研究置于灰区。 **看错或看偏的方面:** 1. **GPT-4参数数量**:所传“100T参数”严重错误,实际约为1.8T。 2. **LLM的数学能力**:诊断“数学差需外挂工具”正确,但“不可能自行学会”的结论被后续“推理模型”在IMO夺金证伪。 3. **价值捕获**:预测价值在应用层,但现实中算力层(如英伟达)捕获了最大利润,模型层反而亏损。 4. **版权问题**:判断“生成物可能无法登记版权”正确,但认为能“规避侵权”则错误,已出现巨额侵权赔偿案例。 5. **信息茧房与世界大同**:机制上AI会将观点向众数平均,但AI正快速走向“千人千面”的个性化,可能制造新茧房而非消解极化。 6. **大模型成本**:预测“局部战争”和玩家涌入正确,但“5-10亿美元封顶”的成本估算与前沿训练实际花费严重不符。 **总结规律:** 1. 判断**方向和机制**比具体数字、程度更可靠。 2. 倾向于**高估短期变化速度,低估长期能力上限**。 3. 容易忽略**问题内部的分布差异**(如总量正确但部分群体受损)。 4. **留有餘地、分层表述**的判断更经得起时间检验。 5. 一些根本性争论,**三年时间不足以给出最终答案**。 这次回顾表明,在快速变化的领域,把握大方向比追求精确预测更重要,同时需警惕绝对化表述并承认认知的局限性。

链捕手1 小時前

三年之后:回看 2023 年我对 ChatGPT 的判断

链捕手1 小時前

交易

現貨
合約
活动图片