The Illusion of Understanding

It’s easy for me to feel informed by high-level summaries, especially those generated by AI. I can quickly skim articles and think I’m up to speed, but these summaries often lack the depth and nuance needed for real understanding. When these summaries are widely syndicated, incorrect assumptions take on an undeserved authority, and I find myself accepting without question. The problem gets worse when these summaries are cited as evidence, further distancing us from the original material.

Why AI Summaries Can Mislead

AI-generated summaries are built for speed and convenience. Unless I specifically prompt tools for critical evaluation, they rarely highlight methodological flaws or contextual limitations. The result? I end up making assumptions based on oversimplified information and missing key points that are present in the primary sources.

Case in Point: AI’s Impact on Jobs

A recent study about generative AI’s effect on jobs is a perfect example. At first glance, the headlines suggested an immediate threat to workers, especially women. But when I dug into the original study, I found a lot of assumptions baked in. For starters, the research was conducted in just one country, Poland, yet the conclusions were extrapolated globally. Respondents were asked to self-evaluate the impact of AI on their jobs, even when their expertise was limited. There were also clear gender and occupational disparities in the roles surveyed.

This is a problem. Unrepresentative samples, limited respondent expertise, and global extrapolations mean the study’s conclusions should be viewed critically. The source material actually acknowledges these limitations and calls for further study. If I had just skimmed the headline and reposted it, I’d be jumping to conclusions and missing the fact that, while not definitive, this topic deserves much deeper investigation.

What I Do Instead

Consult Primary Sources

To avoid making incorrect assumptions, I make a point to consult primary sources whenever possible. Not only do I get a more accurate understanding, but I can also evaluate the evidence, context, and limitations behind any claim. If I truly care about a subject, it’s worth drilling down to find the original study or data.

Skim, But Don’t React

Summaries and lists are great for a quick overview, but they’re often designed to provoke an emotional reaction. If I take a post at face value and respond immediately, I risk lending credibility to misinformation and unwarranted assumptions. Instead, I use summaries to get a high-level overview, then selectively dig into the items that seem most important. If I can’t find any primary sources, I assume the article is probably B.S., even AI-generated content has a source if you look hard enough.

Better AI Summaries

A lot of sites and services claim to provide insightful AI summaries. Over the past year, I’ve tried many, and most are limited by the quality and depth of their source data. Worse, some tools only summarize headlines and publicly available summary articles, creating the illusion of being informed when, in reality, I’m just being marketed to.

Rather than rely on AI to aggregate sources, I subscribe to a selection of newsletters and RSS feeds that bring fresh articles. These are mostly human-curated, and when I notice a headline appearing across several sources, I take it as a sign of importance. But I’m busy, so how do I dig into the details?

Digging In

In short, I start clicking. Before I dive in, I often use Perplexity as a first-pass summary tool. This isn’t a Perplexity advertisement, paid or otherwise, just my opinion of the best tool for the job right now.

Perplexity’s research feature stands out because it breaks down queries, conducts dozens of targeted searches, and evaluates hundreds of sources in real time. By prioritizing authoritative and reputable sources, it generates summaries with citations to the original material.

Another advantage I’ve found is that I can upload the original study (as a PDF) and ask Perplexity to research and summarize it further. This feature is incredibly useful, as it not only condenses long documents but also cites specific sections of the source material.

Takeaway

The world is overflowing with quick takes and AI-generated summaries, and I’ve learned that real understanding comes from digging into original sources and questioning surface-level information. Taking time to consult primary material, to avoid the trap of forming assumptions, has greatly improved my understanding for complex and nuanced topics. In order to be truly informed I need to slow down, do the research, and let the facts speak for themselves.

For the better the Internet continues to be a place of discovery, entertainment and shared zeitgeist. At it’s worst it is a terrible conveyer of current events, of share in-depth analysis and a creator of civil public discourse. Personal posturing and AI generated content may have irrevocably broken things.

What Still Works

Short-Form Video

The original TikTok format struck a nerve making the format work. TikTok wasn’t the first, but its algorithm was successful at serving up and endless stream of engaging content in rapid succession. It was incredibly satisfying and mildly addictive.

Critics say that it shortened attention spans, but that trend was already underway thanks to microblogging and sound bite style news reporting. The game changer was TikTok’s algorithm and backlog of content which kept us scrolling for hours.

Instagram is largely displacing TikTok with established content creators but you can still enjoy discovering random creators via the algorithm. It’s hard to say why creators shifted but it could have been for a number of reasons like the prevalence of the TikTok shop, the unpredictable nature of the algorithm or the questionable decision to start promoting longer videos.

Long-form Video

YouTube tends to dominate in the longish 5-20-minute content video. Automatic video chapters makes it easy to skip around or back to important parts. Videos are typically more polished, but an increasing number feature content creators simply sitting in front of a camera or sharing their screen. Podcasts are also very popular with some saying that it surpasses rival Spotify.

Long-form writing

Substack didn’t invent the newsletter, but the platform made it accessible at a time when everyone was weary of short-form writing and coincidentally had ample time on their hands. More importantly SubStack makes social discovery and sharing easy via Notes. I’ve found, and supported, so many unique authors over the the years.

Notes, a way to recommend content

Short form writing

Most types of short-form writing and threaded comments are mostly NOT useful. The implosion of Twitter scattered that platforms bad behavior to other places like Mastodon, Bluesky and Threads. There is little joy in doom scrolling these sites and worse they encourage a type of slacktivism that is neither effective nor healthy.

“Professional” microblogging on LinkedIn is, at best, a cross-congratulatory mechanism and platform for shameless self-promotion. At worst, it’s a great way to tank a career by posting less than professional hot takes on a troll post.

Broadcast Channels, a direct line to followers

The lone exception is Instagram Broadcast Channels which allows content creators to send messages directly to followers. No public replies are allowed to it is a clean way to communicate short posts directly with subscribers. I’ve joined several during sporting events and found it a pleasant way to stay up to date without the reply spam.

Where Machines Can’t Go (Yet)

AI content generation will continue to overrun quick, easy to produce content. Microblogging, as a format, is mostly dead and engagement numbers will quickly indicate you are shouting into the void.

Quick hit video content continues to thrive but it need to be direct, to the point and eye catching. A teaser cover with outlandish Buzzfeed style titles seem to work the best. Less polished videos easily go viral and platforms like TikTok and Instagram make it easier to record a quick video. Long form video, on the other hand, with a big editing demand is so low ROI that many are relying on AI video editing and narration tools to create content, and it’s ruining the format.

Writers with something to say might want to start a newsletter, but these days everyone has a Substack. To the platform’s credit, it does make accessible the features to manage posts, subscribers and even export subscribers in the event you outgrow the platform. Chances are you won’t, but good writers can sometimes cultivate a small following of highly engaged readers.

Reasonably Human

I base these opinions on my own experience as a media consumer. I’ve created a few meme videos and write a weekly substack of curated interesting reads, but I am not a professional content creator. The examples I have given are from personal experience and listening to the content creators that I enjoy and support.

I choose to create a small amount of content for the lulz, but also in an attempt at credibility for being a real human. In the rare online interaction I always check the account first. If it is blank or seems overly machine generated I’ll refrain. Bots are everywhere and serve to only farm engagement or spark flame wars. It isn’t fun and it isn’t worth it.

Likewise when I leave a comment or reshare I want others to see me as a real human. In 2025 being perceived as a good human is one of the few things we have left.

Something we can all a strive toward

The bubble repeated, for different underlying speculative reasons but effectively the same boom-bust cycle. What happens next to employment and the broad economy is likely a replay of the early 2000s.

Prithvi Raj Chauhan does a great job covering the technical aspects in Dot-Com Bubble 2.0: How Did We Get Here?, I’d like to add some perspective from the employment.

As a Tech Worker

Technologists who remember the furor and hype of the late 90s Dot Com bubble probably recognized the new hype cycle of Bubble 2.0. What really freaked us out was when we started to see the cracks emerge and eventually the hyped companies begin to fall. We tried to warn our younger colleagues who had no professional memory of the time.

I see a lot of similarities between the tech bubbles. If history repeats, it’s worth keeping in mind three distinct phases that we lived through. Keep in mind, by the end of the cycle new tech companies and technologies were born.

Phase 1 (2000-2001)

The first phase was a crisis of confidence. Years of burning investor money with the sole goal of growing let companies into excess. Wall Street convinced retail investors, who had just gained access to platforms like E-Trade that profitability didn’t matter, until it did. When Dot Com companies realized that the next round of funding wasn’t coming, or it would be smaller, they began to panic and laid off thousands.

It happened fast and it was like contagion. Companies that needed to reduce costs for survival laid off first, and then everyone else got nervous and followed suit. Bankruptcies started but didn’t really accelerate until after the stock market had fallen. Office equipment and server racks were on the sidewalks as companies could no longer pay rent and were evicted.

During the initial phase many were laid off in waves. The first were often able to receive severance, but later waves found themselves without a safety net and worse their stock / stock-options worthless. During this period F'dCompany was founded as a site for employees to explain why their companies were going out of business.

Phase 2 (2001-2002)

The second phase unfolded began in 2001 around the time of Sept 11th and quickly deteriorated into a broad economic recession. Here layoffs spread beyond technology and most companies instituted a hiring freeze. This was especially challenging for new graduates who faced a vastly different employment picture in the fall of 2001.

Fraud was a major undoing in several industries including Enron (energy) and WorldCom (telecommunications). The flameouts were rapid, and I remember the day my WorldCom calling card stopped working, only to find out hours later than the company was defunct. People feared terrorism and were afraid of losing their jobs.

Rounds of layoffs continued, but savvy companies made large cuts immediately. If you were unlucky your company made a series of smaller, more frequent rounds of layoffs increasing your personal existential dread. This reached a peak by the end of 2002.

Phase 3 (2002-2004)

The third and final painful chapter began in 2002 and ran through sometime in 2004. People continued to lose their jobs, and businesses continued to close their doors. Unemployment peaked in June 2003 at 6.3%. During the later stage US manufacturing companies began to succumb to a decade of globalization, closing factories or simply going out of business.

Underfunded pensions were a huge liability for older manufacturing companies. In my area Pillotex Corporation famously went bankrupt and dissolved their pension plans that had supported retired workers for years. In other industries pensions were frozen and replaced by 401Ks forcing more of the retirement burden on employees.

During this time getting a job was difficult but not impossible. Work experience was certainly a plus and new graduates continued to struggle to find entry level positions. During this time H1B workers and offshoring became a concern, and many jobs were shipped to lower cost countries like India (tech) and China (manufacturing). In some cases, laid off employees were expected to train their replacements.

In the aftermath interest rates remained low, borrowing was easier and new companies were born. Homeownership became easier, but eventually lead to the 2008 mortgage crisis which was much worse recession in hindsight. Offshoring would accelerate and reach a peak during the next recession.

Takeaway

First, I am not implying that a second, worse recession will automatically follow this one. Each cycle is different, but the outcome (and advice) is always the same. Don’t overspend, save money and be smart about who you work for.

If your employment deal seems too good to be true, it probably won’t last. If your company isn’t profitable and depends on someone or something else for money be wary. It’s important to target working for companies in boring but profitable industries. As one VC put it “prioritize survival”.

Above all chase work experience, not titles or money. No one cares about your title, and money often comes with strings attached. Experience you take with you, and no one can take it from you. And above all trust the experience of your mentors and older colleagues who have lived through this. You’ll get through it, we always do.

Originally posted on BrianWAustin.com