After experiencing the baptism of "waking up every day is a new AI world" in mid-March, humans who took a little breath gradually formed two camps: one was busy calling for the suspension/ banning of large model training, while the other was busy with research on the development of AI technology, vertical field applications, commercial landing forms, and human-computer co-existing issues.
The ZhenFund investment team continued to sort out interesting new AI products, and summarized the latest views and thoughts on AI from technical experts and entrepreneurs in various fields.
This issue will bring to you:
- - 23 products - including 5 new products from big companies and established startups, 9 useful ones, 4 interesting ones, and 5 small products built by Chinese developers;
- - 6 points of view - including a timeline of events calling for the suspension of model training, as well as the thoughts of different technicians, entrepreneurs, and writers on vertical fields and attitudes towards AI. Here, we will not judge the merits of these positions, but objectively summarize and state both sides' behaviors.
It should be noted that given the many explorations of technological development, vertical fields, and human-machine symbiosis work published in the form of papers last week, such as BloombergGPT in the financial sector, ChatDoctor in the medical field, HuggingGPT for tool use, and humans and AI in natural selection, we will share these papers with you in subsequent articles this week!
AI Products
New moves by established companies | From Big
Microsoft
MicrosoftSecurity Copilot
Microsoft, which just turned 48 on April 4, is fully embracing AI and integrating it into its major product lines to establish itself as a leader in the next era, this time with the network security product Security Copilot .
I remember a few months ago when some startups gradually began to use Gen AI to solve the security attack problem, and now I can't help but worry about them, because the giants are making their own strong strides into this field. The reality is that Microsoft's AI Security Copilot is able to simplify emergency understanding of attacks by providing actionable recommendations in minutes, revealing threats, and even predicting attackers' most likely next moves , which could be a breakthrough in cybersecurity . An important advance.
Link - https://www.microsoft.com/en-us/security/business/ai-machine-learning/microsoft-security-copilot
Loop
Microsoft has been experimenting with new office document formats, and in 2020 they introduced Fluid and now Loop, a collaboration tool that is separate from but closely related to MS Office. However, I think many people will find it at a glance - it is too similar to Notion's UI, almost exactly the same, the same text blocks, pages and workspaces - but after a day of trial, I found that its smoothness is worse than Notion Far.
Still, it's a noteworthy move, as it's also competing to some extent with Google Workspace, which currently only has a web-based app, but will soon have an app. Also worth mentioning is that it has built-in AI Copilot.
Link - https://www.microsoft.com/en-us/microsoft-loop
Ghostwriter by Replit & Google
On the defensive side, we have OpenAI and Microsoft, they have the omniscient, omnipotent GitHub Copilot, and now we have new siege players: Replit and Google. Replit is upgrading their recently released programming assistant Ghostwriter. This time, they chose to cooperate with Google and use their superior language model. It is no longer a temporary copilot, but an end-to-end project assistant that can write code, fix bugs, make suggestions, write tests, deploy, review and help collaborate... it's up to you!
Link - https://replit.com/site/ghostwriter
Video-to-3D API by Luma AI
In the previous Newsletter, we introduced the amazing 3D real-time rendering technology launched by Luma AI. Now, through their Video-to-3D API , we can further realize the "Internet-scale 3D world" - no Wrong, it's the API service - for only $1 per scene (of course, this may not be cheap), developers can get NeRF modeling services, input a video, and be fully immersed in the interactive 3D scene !
Link - https://captures.lumalabs.ai/luma-api
Vincent presentation by Tome & Gamma
To be honest, I'm a bit tired of the same business productivity tools that look different but are actually pretty much the same.
But from a $26 million Series A round last year to a $43 million Series B round in February this year, Tome is an impressively fast-growing startup. On March 23, they launched the first document-based AI presentation tool , based on GPT-4, capable of converting documents into presentations, narratives, and stories. Combined with Tome's self-developed rendering system, this is not only another tool for generating slideshows, nor can it be simply compared with Powerpoint integrated with GPT. The smooth generation process is enough to become their competitiveness!
Immediately after last Friday, Gamma also launched its own Wensheng presentation tool , which has a different style and the same smooth workflow!
Link - Tome: https://beta.tome.app/Link - Gamma: https://gamma.app/
Useful｜For Money
Fixie.ai
Founder Matt Welsh described Fixie.ai, which has just received $17 million in seed round financing: "The first PaaS company to build an LLM experience centered on the enterprise" - but One AI, which received financing at the end of May 2022, may not think so.
In short, Fixie.ai is an LLM-driven agent that interacts with external systems . It aims to link LLM with enterprise data, systems and workflows. Currently, it can communicate with databases, APIs (such as GitHub), productivity tools ( such as Google Calendar) and public data sources such as search engines and social media to generate and process images and text.
In addition, the team at Fixie.ai deserves our attention:
- CEO Matt Welsh spent nearly a decade as an engineering lead on the Google Chrome team;- CPO Zach Koch, former Director of Product at Shopify and former Google Chrome and Android team lead;- CTO Justin Uberti was one of the original architects of AOL Instant Messenger;- Chief AI Officer Hessam Bagherinezhad was previously Apple's head of machine learning, responsible for products such as the iPhone and Apple Watch.
Link - https://www.fixie.ai/
Regex.ai
An AI-based automatic regular expression generation tool , out of the box, what you see is what you get, select data, you can generate regular expressions, and provide a variety of data extraction methods, compared with the previous generation for debugging regular expressions Tools such as Rubular have taken a big step forward, and compared to ChatGPT, they are also more intuitive - no need for language description, directly select the part you want to get, and automatically complete the regular expression.
Link - https://regex.ai/
Enhance AI
In one sentence: low/no-code tool to integrate state-of-the-art LLM (currently OpenAI GPT model) in-product within a minute!
Link - https://www.enhanceai.dev/?ref=producthunt
Reimagine by HomeByte
The automatic rendering of interior decoration is undoubtedly one of the most intuitive landing scenarios for visual models, and the popular house sales platform HomeByte has integrated this function into the platform and opened it to C-end users-when buying a house, you can imagine future living the way you go in!
Link - https://homebyte.com/reimagine/tag/latest
Playground v1 by Playground AI
Say hello to Playground v1, a new and awesomely generated visual model!
According to Playground AI founder Suhail, the model can improve the overall HDR quality of generated images. I must say that the resulting pictures look really sharp and very detailed. Currently, users can try it for free on the Playground official website (although sometimes there may be small bugs), and there are many novel styles to choose from!
Link - https://playgroundai.com/
Clip Interrogator
"Picture-generated text" reverse tool - when we saw a favorite AI-generated picture, we always had to find the author and ask what the prompt is. Now, we only need to upload the Clip Interrogator! Actual uploading of real photos can also generate a corresponding prompt~
Of course, for Midjourney users, there is no need to be so troublesome. In the early hours of yesterday, David Holz announced that Midjourney has officially realized the "left-handed and right-handed" and "picture text" reverse output functions, and it is online!
Link - https://replicate.com/pharmapsychotic/clip-interrogator
GPTCHA
The increasingly powerful AI has already caused us to worry about fraud automation, and CPTCHA, which just won the first place in the AI for Good Hackathon, is committed to solving an important part of it-telephone fraud. It's a GPT-4-powered gadget built by three developers that intercepts suspicious calls and chats with the caller with a virtual voice until the call is confirmed to be legitimate and safe.
Link - http://gptcha.ai/