Time to do a bit more AI, considering how many announcements came in the past 48 hours.
OpenAI announced on Tuesday GPT-4, the latest iteration of their large language model. This is multimodal, accepting text and image inputs to generate text outputs. The impressive demo included inputting the US tax code and answering questions, scanning a drawing of a website on a napkin and generating the code for that website, and describing in detail what appeared in images.
In the announcement. OpenAI claimed that GPT-4 scored in the 90th or higher percentiles for the bar exam, the verbal GRE, and the reading and writing portions of the SAT—without specifically training the model for these tests. This means an automated system took these exams without “studying” and could score higher than most humans. According to the company, it could achieve 1410 on the SAT, pass the Bar and GREs, and scored all fours and fives on AP Art History, Biology, Calculus BC, and Chemistry exams—high enough to get college credit.
And the company claims GPT-4 is 82% less likely to respond to requests for disallowed content and 40% more likely to produce accurate responses than GPT-3.5 on our internal evaluations. The company released it as closed source, breaking with prior tradition, not including details about the architecture (including model size), hardware, training compute, dataset construction, and training method.
Want to try GPT-4 now? Microsoft announced that the Bing AI bot has already been running GPT-4. And the company has removed the waitlist for Bing Chat, making it available to anyone. The company also is rolling out the Bing AI chatbot as a sidebar in its Edge browser, available now.
Microsoft has announced Copilot for its Microsoft 365 apps. Powered by GPT-4, it sits alongside the Microsoft 365 apps as a chatbot to generate text in documents, create PowerPoint presentations based on Word documents, or even help use features like PivotTables in Excel. Currently being tested with about 20 customers, the intention is to expand the previews in the coming months and share pricing and details along that timeframe.
The rollouts continue – LinkedIn introduced AI-powered writing suggestions, initially focused on improving profiles and writing job descriptions.
Google-backed Anthropic has launched its chatbot, Claude. Anthropic says its early clients report the tool’s “less likely to produce harmful outputs” and is “easier to converse with.” Google has opened up its AI model PaLM, intended to have an offering to compete with OpenAI or Meta’s offerings. And the company showed off its medical-focused one, a chatbot called Med-PaLM 2, which consistently passes medical exam questions with a score of 85%, placing it at the “expert” doctor level, the company said.
Of course, it’s not all good — Security experts have developed a method for tricking the well-known AI Chatbot ChatGPT into creating malware. Quoting Cybersecurity News, CodeBlue29 used OpenAI’s ChatGPT to create a ransomware test sample to test different EDR solutions to help them decide which product to purchase for their company.
They were able to have ChatGPT generate pieces of code that were able to append together to create a working sample of custom ransomware in Python despite having little programming experience.
During the testing of the ransomware against several EDR solutions, the malware was able to circumvent one of the vendor’s defenses. Codeblue29 reported the discovery to the EDR vendor via their bug bounty program, which resulted in the issue being resolved.
Finally, let’s leave a big idea to ponder – could AI replace the CEO? NetDragon Websoft — a Hong Kong-based online gaming firm with $2.1B in annual revenue – did just that, and the company has outperformed Hong Kong’s stock market. I’ll link to the Hustle analysis here – because further data show how CEO pay isn’t correlated to performance… with lower-paid CEOs outperforming their higher-paid counterparts.
Why do we care?
Will the CEO be automated? Not really – but they may be augmented by AI or have their roles further dispersed.
While this show won’t become the Business of AI, this is the most exciting tech area right now, full stop. Context — ChatGPT hit its first million users in record time, taking just five days to reach that number. Jan 2023, it hit 57 million and 100 million in February. This is Shadow IT on speed.
The tools are in users’ hands. Putting it right in the browser drives adoption… and say it again, drives adoption without IT intervention. Without the bosses knowing. Huge space here to consult and help businesses be effective and reduce exposure. Models and prompts are the first areas to get educated on.
Be First to Comment