长推:对红杉资本《The New Language Model Stack》文章的8点分析

MarsBitОпубликовано 2023-06-16Обновлено 2023-06-16

Введение

干货非常多。很多内容对AI创业者应该会有不少帮助

注:本文来自@FinanceYF5 推特,其是@33daoweb3 成员,Gofans顾问,原推文内容由MarsBit整理如下:
红杉资本6月14日发布最新文章《The New Language Model Stack》
访谈了投资Portfolio里的33家小到种子轮,大到已经上市的公司后总结出来的。
全文总共有8点分析,干货非常多。很多内容对AI创业者应该会有不少帮助
Thread:

AI人工智能


1.几乎所有公司都在产品中用到语言模型
几乎每一家公司都在将语言模型构建到他们的产品中
- 从代码和数据科学副驾驶,到为客户、开发者、员工和纯娱乐提供的聊天机器人。
许多公司正在用AI优先的视角重新设想整个工作流程。
这些只是几个例子,而且只是个开始。
2.这些应用主要基于API、检索和编排,但开源使用也在增长
-65%的公司已经将应用投入生产
-94%的公司正在使用基础模型API
-88%的公司认为检索技术
-38%的公司对像LangChain的框架感兴趣
-15%的公司从头开始或使用开源资源构建了自定义的语言模型

AI人工智能


与每一位实践者交谈的结果显示,AI的发展速度太快,以至于很难对最终的技术栈有高度的信心,
但大家一致认为,LLM API将继续作为一个关键支柱,
其次是检索机制和像LangChain这样的开发框架。
开源和自定义模型的训练和调整似乎也在增长。
语言模型技术栈的其他领域也很重要,但成熟度较低。
3.公司希望根据自己的上下文来定制语言模型
目前,有三种主要的方式来定制语言模型:
从头开始训练自定义模型,难度最高。
微调基础模型,难度中等。
使用预训练模型并检索相关上下文,难度最低

AI人工智能


4.如今API调用和训练模型是相互独立的,但未来两者将慢慢融合在一起
我们可能会觉得我们面临着两种技术栈的选择:
一种是利用LLM API的技术栈
另一种是训练自定义语言模型的技术栈
越来越多的公司对训练和微调自己的模型产生了兴趣。随着时间的推移,LLM API技术栈和自定义模型技术栈会越来越融合
5.技术栈正在变得越来越易于开发者使用
LangChain通过抽象化常见问题,帮助开发者构建LLM应用:
-将模型组合成更高级的系统,
-将多个模型调用链接在一起,
-将模型连接到工具和数据源,
-构建能够操作这些工具的代理,
-以及通过简化语言模型切换的过程,帮助避免对供应商的依赖。
6.语言模型需要变得更可靠(输出质量、数据隐私和安全性)
许多公司希望有更好的工具来处理数据隐私、隔离、安全、版权和监控模型输出。
来自金融科技到医疗保健的受监管行业的公司特别关注这一点
随着政策的明确和更多的安全措施到位,语言模型将得到更好的信任
7.语言模型应用将变得越来越多模态
8.目前还仅仅只是初期阶段
AI刚刚开始渗透到技术的每一个角落,只有65%的受访者今天已经开始尝试,而且其中很多都是相对简单的应用。
基础设施层(我理解就是中间层的意思)将在未来几年内快速发展
原文:
https://twitter.com/michelle_fradin/status/1669117628521271298

Похожее

Claude Code Introduces Dynamic Workflows: Enabling AI to Form Teams and Collaborate

Claude Code introduces dynamic workflows, enabling AI to coordinate teams of specialized agents for complex tasks. This transforms Claude from a code assistant into a programmable workbench. Workflows address key limitations of single-agent systems: agentic laziness (premature task completion), self-preferential bias (favoring own outputs), and goal drift (losing sight of original objectives). The system allows Claude to dynamically create execution frameworks using JavaScript. It can split tasks, dispatch parallel agents for isolated work (e.g., in separate worktrees), implement adversarial validation, run tournaments, and synthesize results. This multi-agent approach is valuable for tasks requiring deep research, factual verification, code migration, root cause analysis, large-scale triage, and qualitative sorting. Key patterns include: classify-and-route, fan-out-and-synthesize, adversarial verification, generate-and-filter, tournaments, and loop-until-done. While token usage is higher, workflows excel where tasks resemble programming—needing problem decomposition, isolated context, hypothesis testing, and handling many details. They extend Claude Code's utility beyond technical work to areas like business plan review, resume screening, and naming brainstorm. The feature is not a universal solution but points to a future where AI tool competitiveness depends on organizing reliable, reusable, and auditable execution flows for complex goals.

marsbit37 мин. назад

Claude Code Introduces Dynamic Workflows: Enabling AI to Form Teams and Collaborate

marsbit37 мин. назад

Hyperliquid, Wall Street's 24/7 Trading Convenience Store

Hyperliquid: The 24/7 Trading "Convenience Store" for Wall Street Hyperliquid, a decentralized cryptocurrency exchange, has become a go-to platform for Wall Street traders seeking to trade around the clock, especially during traditional market closures. Founded by Jeff Yan, a former quantitative trader, after the FTX collapse, the platform emphasizes user self-custody of assets. It offers a wide range of perpetual contracts—leveraged derivatives with no expiry—on assets from Bitcoin and crude oil to the S&P 500 and even pre-IPO companies like SpaceX. A notable example involves a hedge fund trader who capitalized on geopolitical news over a weekend, securing a 243% return on oil derivatives before markets reopened. The platform, run by just 11 employees, generated approximately $800 million in revenue last year, and its native token HYPE has seen significant growth. Its rise highlights the merging of traditional finance and crypto. While U.S. users are currently restricted, recent CFTC rule changes could open access. The platform is known for its transparency, having processed $10 billion in liquidations during a market crash while competitors faltered. Regulators warn of the high risks and complexity of perpetual contracts for retail investors. Key to its appeal is a strong community culture, direct engagement with founders, and a simple interface. Despite rules against VPN use, it attracts global users with its permissionless approach. Hyperliquid plans to expand into prediction markets and options, aiming to eventually host all financial activity.

marsbit37 мин. назад

Hyperliquid, Wall Street's 24/7 Trading Convenience Store

marsbit37 мин. назад

Who Funds the Agents?

**Summary: Who Funds AI Agents?** OpenAI recently shut down a feature allowing AI agents to shop for users, highlighting the challenge of creating a secure and regulated environment for agent-driven transactions. While payment infrastructure exists, a crucial governance layer—defining spending limits, fraud detection, tax handling, and return policies—is largely missing. The potential is enormous: AI agents already processed $73M across 176M transactions last year, with McKinsey forecasting this could grow to $3-5T in global consumer commerce by 2030. The core competition isn't just about processing payments, which can be very cheap (especially with crypto-based settlement), but about controlling the rules that govern agent spending. Key players like Stripe and Coinbase are racing to dominate this governance layer. Stripe's acquisition of wallet provider Privy allows it to set spending policies, identity checks, and human-in-the-loop approvals directly at the wallet level. Similarly, Coinbase's stack, including its x402 protocol and AgentKit, embeds governance rules. This vertical integration across settlement, wallet, and governance layers is becoming the dominant strategy. Control over the governance layer is where significant future value lies. If agents handle trillions in transactions, even a small fee for managing compliance, fraud prevention, and policy enforcement could generate billions in annual revenue. The companies that successfully integrate across the payment stack will capture value from idle agent balances, transaction fees, and governance services, positioning themselves as the foundational banks of the AI agent economy.

marsbit1 ч. назад

Who Funds the Agents?

marsbit1 ч. назад

Торговля

Спот
Фьючерсы
活动图片