- The Multidisciplined
- Posts
- 🤖 ChatGPT can now use the Internet
🤖 ChatGPT can now use the Internet
OpenAI revving up once again
Lorel Sim & Reily Haag
October 03, 2023
The TLDR
Highlight: Always pushing boundaries, OpenAI is rolling out several new game-changing features that support a more connected experience. Today, we’re reviewing ChatGPT’s new Internet Browsing feature and sharing 4 practice prompts
Musing of the Week: On the heels of a resolution between the WGA and the AMPTP, Tom Hanks denounced a dental plan ad that used his likeness to target consumers. Is Hollywood itself really the biggest thing that talent needs to worry about?
💣 A Major ChatGPT Update
The AI newsfeed is constantly buzzing with ‘new feature’ announcements from big tech companies, all in a bid to stay relevant in the AI arena. OpenAI’s ChatGPT arguably maintains an advantage in performance and public opinion, but a major competitive weakness is its lack of pre-existing integration into our daily lives, unlike Google, Microsoft, Meta, and Amazon (which are all vying for leadership in AI). That said, OpenAI is clearly racing toward a connected, sticky, embedded experience and just teased several new features.
🚀 New ChatGPT features on the horizon for Plus & Enterprise:
Multi-modal capabilities (e.g., image input, text-to-speech)
Improvements to DALL-E (more realistic human hands and letters/numbers)
Real-time Internet browsing
Multi-modality and DALL-E 3 are still being rolled out over the next month, so today, we’ll focus on reviewing the new Internet Browsing plug-in.
🤨 (Real quick - why are features rolled out gradually?)
OpenAI states: “We believe in making our tools available gradually, which allows us to make improvements and refine risk mitigations over time... This strategy becomes even more important with advanced models involving voice and vision.”
👩💻 Let’s talk about ChatGPT + Internet Browsing
Internet Browsing unlocks more data
From launch, the biggest disclaimer about ChatGPT was that it only had knowledge of events up to September 2021. This meant that users had to upload their own context for post-2021 information, fact-check heavily, and apply a grain of salt to the output.
This is actually attempt #2
Internet Browsing was introduced back in May 2023, but OpenAI rolled it back after users started wielding it unethically. After the rollback, users could still try accessing the Internet within ChatGPT conversations by using the less reliable WebPilot plug-in.
How to use Internet Browsing
As mentioned, Internet Browsing is a plug-in that is available for Plus and Enterprise users. Login to ChatGPT, start a new conversation using GPT-4, and turn on “Browse with Bing (Beta).”
OpenAI is refining these models over time, and Internet Browsing might not be able to handle your most robust prompts yet. We tested out a bunch of prompts - here are 4 great use cases that Internet Browsing can handle. Copy our prompts below to practice!
[Prompt for problem-solving]
Hi Chat, the Airtag in my luggage isn't working anymore. Can you search the Internet for all possible issues and help me troubleshoot?[Prompt for fact-checking]
Hi Chat, I just read that the US-based ecommerce industry is at $1.03 Trillion in 2023. Is that accurate? Please review the Internet and fact check the number I gave you with 10 different reputable sources.[Prompt for interview practice]
Act as an interviewer and help me prepare for a first-round interview for the role of General Counsel with Leyton, a global business consultancy. Review the job description at this URL (https://builtin.com/job/legal/general-counsel/2018464) and begin my interview. Conduct a realistic conversation, only asking me one question at a time and asking follow-ups if you deem necessary.[Prompt for factual research]
I need to find 5 really interesting, obscure facts about the coffee industry for a school paper. Please search the Internet and present 10 options for me to use.^This last prompt for coffee facts actually yielded some awesome trivia. For our coffee nerds, click here for the output.
✍️ Notes and call-outs for Internet Browsing
Website owners can choose to block ChatGPT’s web crawler, so you might not be able to access all websites from ChatGPT
Even if you have the Internet Browsing plug-in turned on, you need to prompt it to use the Internet. Include something like “Review this URL” or “Search the Internet for __.”
ChatGPT doesn’t automatically tell you the search terms it’s querying. If you want to know, you’ll have to ask it.
We love that ChatGPT can cite its sources now. It makes it way easier to pull certain threads and check facts.
🧠 Musing of the Week
Last week, the Writers Guild of America won a landmark victory negotiating protections against AI into their new union contract. This sets a new precedent for negotiation around the use of AI in industry and will doubtlessly help SAG-AFTRA to secure similar protections for their own members. Without question, this is a step towards improved standards around the use of AI.
However, only a few days after the end of the WGA strike, AI was back in the news as, within the span of 24 hours, Tom Hanks and Mr. Beast were both forced to make statements denouncing endorsements using AI versions of their likeness. This brings to light a broader question about the impact of these new contracts.
Both WGA and SAG-AFTRA contracts are negotiated with the Alliance of Motion Picture and Television Producers (AMPTP), a collective bargaining organization representing the broader entertainment industry. While new contracts regulate existing studios, they have no impact on organizations or individuals that exist outside of the AMPTP.
As we approach the point where every individual with a laptop and a marginal knowledge of AI tools can produce a believable representation of anyone with a digital footprint, it will become incredibly hard to differentiate the real from the noise. Though random people and companies using celebrities likenesses is likely illegal in most jurisdictions, at scale, this becomes an issue of enforcement over legality. Without drastically improved mechanisms, it’s difficult to imagine tight enough enforcement to prevent the proliferation of these reproductions. This presents a clear and present danger to the way that we consume and value content that will require careful reconsideration of our values and priorities. From where we’re sitting this seems like the beginning of the road and it’s likely that this will need to become a central point of discussion over the coming months.
🙌 If you’re hyped about the generative AI industry specifically, here are some of the coolest roles we’ve seen this week:
🔨 Check out these other AI tools we’ve been looking at this week:
That’s all for this week. See you next Tuesday!
Lorel & Reily