我们能否制造出比人类更聪明的更小的开源LLM模型?
[
[
I am Harish SG, a security researcher who studied Masters in Cybersecurity at UT Dallas and AI security engineer at Cisco, previously hunted on the Microsoft Bug Bounty Program and Google VRP.
我是Harish SG,一名安全研究员,在德克萨斯大学达拉斯分校攻读了网络安全硕士学位,并在思科担任人工智能安全工程师,之前曾参与微软漏洞赏金计划和谷歌VRP。
I am sharing this article for awareness and educational purposes only and I am sharing only personal opinions and none of these are related to my work at Cisco.
我分享这篇文章只是为了提高意识和教育目的,其中的观点仅代表个人观点,与我在思科的工作无关。
Disclaimer: I am not a AI researcher or expert and I do security research on LLMs . This work is fully based on my understanding of LLMs and its capabilities.
免责声明:我不是人工智能研究员或专家,我在LLMs上进行安全研究。这项工作完全基于我对LLMs及其能力的理解。
This article is focused on my recent AI research on making opensource models to outperform other closed sourced models and making current SOTA (State Of the Art) models such as Claude Sonnet 3.5 to outperform reasoning SOTA model OpenAI O1-preview and O1 mini(both of them has phd scholar level intelligence according to OpenAI).
本文侧重于我最近在人工智能研究中的开源模型,以超越其他封闭源模型,并使当前的SOTA(State Of the Art)模型,如Claude Sonnet 3.5,超越推理SOTA模型OpenAI O1-preview和O1 mini(根据OpenAI的说法,它们都具有博士学者级别的智能)。
What is reasoning in LLM ?
什么是LLM中的推理?
Reasoning in LLMs refers to the ability of these models to:
LLM中的推理能力指的是这些模型具备以下能力:
- Think logically
- 逻辑思考
- Draw inferences
- 推断出结论
- Solve complex problems
- 解决复杂问题
- Make sound decisions based on available information
- 根据可用信息做出明智决策
While LLM are not explicitly trained to reason, they have exhibited behaviors that sometimes resemble reasoning capabilities except O1 and O1 mini
虽然LLM并没有明确接受推理训练,但它们表现出的行为有时与推理能力相似,除了O1和O1 mini之外
Why Reasoning in LLM’s Matters?
为什么LLM的推理很重要?
The ability of LLM to reason is significant for several reasons:
LLM的推理能力对于几个原因都很重要:
- Deeper Understanding: True reasoning abilities would indicate that LLM can g...