š¤ This Week in AI
PLUS: This Overlooked ChatGPT Feature Lets You Personalize and Improve Every Response
Iād love to star by saying āHappy Fridayā but I know this week has been a rough week for many, and my thoughts are with everyone feeling the weight of it.
So, in the spirit of staying grounded, Iām here as usual with my weekly updates and a mini-tutorialā something small yet practical to keep us moving forward.
If youāre tired of explaining the same details to ChatGPT over and over, thereās a simple way to share your key info and preferences once, so it remembers them for every conversation and gives you more relevant, personalized responses. This week, Iāll show you how to do just this with Custom Instructions.
For more frequent AI updates and insights, follow me on LinkedIn and Threads.
But first, hereās what you need to know about AI this week (clickable links appear in orange in emails and underlined in the Substack app)::
How did tech giants react to Trumpās big win? Why tell you when I can just show youā¦
And then there is whatever this isā¦
I guess Elon is now part of the Trump family.
But where is Melania? And I donāt know who that woman is to Elonās right, but I want her outfit!
So, what will a Trump administration likely mean for AI?
Well, he has already promised to reverse President Biden's executive order on AI, which emphasizes safety and security measures. Weāll likely have:
Less regulation and oversight, with faster development aimed at outpacing China by treating AI as a national security priorityāeven if it reduces focus on risks like bias, misinformation, and data privacy.
With support from tech leaders, we could also see closer alignment between government and Big Tech, raising questions about accountability for AIās unintended consequences and social impact.
We should all buckle up. Itās gonna be a bumpy ride.
A new Wharton study shares key insights on the trends, challenges, and impact of Gen AI in business. Here are a few highlights:
72% of executives now use Gen AI weeklyāalmost double last yearās rate, with adoption also nearly doubling in areas like Marketing, HR, and Operations.
Gen AI is proving itself in both analytical and creative work. Top use-cases include:
Document and Proposal Writing/Editing (64%)
Data Analysis and Analytics (62%)
Document/Meeting Summarization (59%)
Marketing Content Creation (58%)
Personalized Marketing and Advertising (58%)
Frequent Gen AI usage is highest among:
Smaller companies (revenue $50M to $250M): 80%
Mid-sized companies (revenue $250M to $2B): 78%
Younger individuals (ages 18-34): 80%
As I mentioned in my rant about AI and Hollywood š last week, small to medium sized companies have an outsized opportunity to gain a competitive edge with this tech.
Key challenges remain: Accuracy, data privacy, and ethical concerns are still factors, though urgency has softened. For many, the focus is now on building AI strategies that can adapt and scale.
The real advantage isnāt in having access to AI technology (itās everywhere, and employees are already using itāoften without permission and on personal devices), but in how creatively and effectively you weave it into your workflows and the operational fabric of your business.
Investing in comprehensive employee training and developing clear guidelines on AI use will go a long way in addressing these challenges.
With the right training, employees can learn when to use AI (and when not to), how to communicate their business problems clearly, choose the right tool for each task, andāmost importantlyāapply their expertise to verify and improve AIās results.
Claude just launched a feature that finally lets users analyze visuals inside PDFs. With its latest model (Sonnet 3.5), it can now see, understand, interpret and analyze images, charts, tables, and even complex infographics within PDF documentsāavailable only to Pro users ($20/month).
There are so many use cases for this capability. Here are just a few:
Analyze financial reports with visual data
Pull data and insights from legal documents or scientific research papers
Translate complex documents with visuals into other languages
Interpret medical documents with images or scans like x-rays
Transcribe and interpret scanned historical documents and maps
Create text descriptions for screen readers to improve accessibility for vision-impaired users
Transform disorganized or image-heavy PDFs into well-structured, formatted documents
ā ļø Limitations: PDFs need to be under 32MB, fewer than 100 pages, and canāt be encrypted or password-protected.
šļø A reminder for ChatGPT Users: ChatGPT still canāt process images within PDFs. So, if youāre working with PDFs that include visuals with critical information, take screenshots of the visuals and upload them alongside the PDF so ChatGPT can analyze them together.
Disney has created a new Office of Technology Enablement to guide the companyās adoption of AI and mixed reality across film, TV, and theme parks. The unit, which is expected to expand to 100 employees, will focus on improving consumer experience while managing risks like IP concerns, a priority as AI raises complex challenges for creative industries
Runway released Advanced Camera Control which lets users control camera movements, including panning, zooming, and tracking shots with adjustable intensity, while keeping 3D depth and spatial relationships consistent throughout scenes. Ā
It looks pretty good. You can see some of the examples š:
Amazon Prime Video introduced a new AI-powered feature called X-Ray Recaps, which creates spoiler-free text summaries for full TV seasons, individual episodes, and even parts of episodes. At launch, the feature will work with Amazon MGM Studios Original series and be available on Fire TV devices, with support for more devices coming later this year.
Iām not sure how helpful a spoiler-free recap for an entire season will be, but I guess weāll see how it works...
Fashion brands like Mango, Louis Vuitton, Nike, and Levi Strauss are among a growing number of companies turning to AI models for faster, more affordable ad production. Mangoās tech team uses AI to build mood boards and designs that align with brand aesthetics, freeing designers to focus on more creative, high-value work.
If youāre a parent who wants to help your kids learn with AIāwithout just giving them answersā the below tip is a great start.
Though sooner or later (likely sooner), knowing how to get the best out of AI will be essential for the future of work.
Meta has changed its policy to allow U.S. government agencies and contractors to use its AI models for national security purposes to help the U.S. maintain a technological edge. This shift is an exception to Metaās āacceptable use policy,ā which bans the companyās software from being used in military, warfare, and nuclear industries.
Meanwhile, Anthropic has partnered with Palantir, a data analytics firm focused on secure defense applications, to deliver its Claude AI models to U.S. intelligence and defense agencies, aiming to improve analysis, streamline workflows, and support complex decision-making.
AI-driven job interviews are quickly becoming the new norm, with companies like Chipotle using virtual recruiters to screen and interact with applicants in real time. While experts say AI makes hiring faster and more objective, some candidates find the process impersonal and worry about potential biases.
The Netherlands' largest publisher, VBK, is using AI to translate a small number of commercial fiction books into English. The process, which excludes literary titles, will involve human editors and require authorsā consent.
Iām so excited for these types of use cases that introduce fresh voices to global audiences.
A new study finds that AI creates clearer, more accessible summaries of research papers than scientists, which improves public understanding and raises trust in scientistsāeven when readers donāt realize AI wrote the summaries.
And if you, like me, are AI nerdy š¤ and want to understand the difference between open-source models (like Metaās Llama) and closed ones (like OpenAIās GPT-4) ā and why it REALLY matters ā this piece from Time is a great place to start.
INSIDER TIPS
This Overlooked ChatGPT Feature Lets You Personalize and Improve Every Response šÆ
If you're using ChatGPT regularly, chances are you've wondered, āHow can I make it remember important things about me and my needs and preferences without repeating myself in every conversation?ā
The answer lies in Custom Instructions, a feature that lets you share information about yourself and your preferences, and that then applies them to EVERY conversation to give you more tailored and personalized responses.
Custom instructions are ideal for setting baseline information or long-standing preferences that donāt often change (field of work, role and responsibilities, interests, preferred tone or response formats).
To create them go to:
iOS & Android: Go to Settings > Personalization > Customize ChatGPT (Toggle āEnable for new chatsā ON)
Web: Go to Settings > Personalization > Custom Instructions (Toggle āCustom Instructionsā ON)
Select it, and you'll get two fields, each with a 1500-character limit:
1ļøā£What would you like ChatGPT to know about you to provide better responses? This is where you can add context about yourself, your role, company, interests or even your longer-term goals. The more relevant details you share, the better ChatGPT can tailor its responses. Here are some examples:
Ā āI lead corporate communication for a global media company focused on crisis management.ā
āIām a small business owner who runs an eco-friendly retail shop. Iām looking for ways to increase customer loyalty, drive social media engagement, and source sustainable products while increasing profitability.ā
Ā āIām a 42-year-old stay-at-home mom who likes to find creative ways to educate and entertain my kids and prepare family meals.ā
Thereās no right or wrong way to respond. The key is to ask yourself: āWhat information would be helpful for ChatGPT to know about me, given the tasks I typically ask it to help with?ā
2ļøā£How would you like ChatGPT to respond? This is where you can give ChatGPT instructions and guidelines about how to youād like it to respond. Anything youād like it to do or not do goes here. Here are some examples:
āStructure your responses in short, digestible paragraphs or bullet points."
āAvoid jargon and fluff.ā
āCite credible sources whenever making factual claims, and include URLs where possible.ā
āAsk follow-up questions to encourage deeper thinking if my initial query seems vague or incomplete.ā
āUse a conversational tone but maintain professionalism.ā
And hereās a favorite of mine: āConvey complex concepts clearly, even to non-technical audiences, while retaining the details and nuances.ā
Hereās a glimpse at my own custom instructions. The screenshot shows only a few lines from each field, but you get the idea!
You can edit, update or delete, or turn this setting on or off at any time. Just revisit the settings to tweak the instructions as your preferences and goals evolve.
Take 15 minutes to think through and set your custom instructions today (or this weekend). Youāll get far less generic and more relevant responses every single time.
In case you missed last weekās post, you can find it š:
That's all for this week.
Iāll see you next Friday. Thoughts, feedback and questions are welcome and much appreciated. Shoot me a note at avi@joinsavvyavi.com.
Stay curious,
Avi
ššš P.S. A huge thank you to my paid subscribers and those of you who share this newsletter with curious friends and coworkers. It takes me about 8+ hours each week to curate, simplify the complex, and write this newsletter. So, your support means the world to me, as it helps me make this process sustainable.