Fonts for rendering lines and bars from data

Google Fonts now provides two open source fonts by Dmitry Ivanov that let you make simplified, small to medium line and bar charts based on data: Linefont and Wavefont. These might come in handy when you want to embed small charts in body text.

Tags: , ,

Manual data labeling behind the AI

One of the things that makes AI seem neat is that it sometimes feels like magic when you enter a question and get a reasonable answer in a human-like tone. For Bloomberg, Davey Alba reports on how some of that magic is just a bunch of people labeling data for low wages:

Other technology companies training AI products also hire human contractors to improve them. In January, Time reported that laborers in Kenya, paid $2 an hour, had worked to make ChatGPT less toxic. Other tech giants, including Meta Platforms Inc., Amazon.com Inc. and Apple Inc. make use of subcontracted staff to moderate social network content and product reviews, and to provide technical support and customer service.

“If you want to ask, what is the secret sauce of Bard and ChatGPT? It’s all of the internet. And it’s all of this labeled data that these labelers create,” said Laura Edelson, a computer scientist at New York University. “It’s worth remembering that these systems are not the work of magicians — they are the work of thousands of people and their low-paid labor.”

Tags: , , ,

Generating music from text

Researchers at Google built a model that generates music based on brief text descriptions:

We introduce MusicLM, a model generating high-fidelity music from text descriptions such as “a calming violin melody backed by a distorted guitar riff”. MusicLM casts the process of conditional music generation as a hierarchical sequence-to-sequence modeling task, and it generates music at 24 kHz that remains consistent over several minutes. Our experiments show that MusicLM outperforms previous systems both in audio quality and adherence to the text description. Moreover, we demonstrate that MusicLM can be conditioned on both text and a melody in that it can transform whistled and hummed melodies according to the style described in a text caption. To support future research, we publicly release MusicCaps, a dataset composed of 5.5k music-text pairs, with rich text descriptions provided by human experts.

I’m not entirely sure I like where this road goes, but the results are impressive.

Tags: , ,

Google Maps incorrectly pointing people to crisis pregnancy centers

Davey Alba and Jack Gillum, for Bloomberg, found that Google Maps commonly points people to crisis pregnancy centers, non-medical locations that encourage women to follow through with pregnancy, when they search for “abortion clinic”.

Tags: , , ,

Interest levels for political issues mapped

To estimate public interest in the many political issues across the United States, Axios used Google Trends data to map issues by congressional district. Switch between the many topics, and you see a choropleth map (that can change to a cartogram), along with a barcode chart to show the distribution of interest among all districts.

I’m not sure if it’s that beneficial to see the overall geographic distributions for most topics, but it’s useful as a point of reference to look at specific districts. For me, the barcode chart is the most interesting with the distributions shifting quite a bit from topic to topic.

Tags: , , ,

New shopping search patterns from the pandemic

Schema Design, Google Trends, and Axios collaborated on The New Normal, looking at how searches for certain products has changed since the pandemic started. Keywords were taken from Google’s product taxonomy, and search volumes are from Google Shopping.

From there, the keywords, compared to search from 2019, were categorized as a new normal, unusual, or about the same as before. They categorized the words manually instead of defining a metric, which surprised me. It seems like it would’ve been useful for sorting beyond alphabetical. Still interesting to poke at though.

Tags: , ,

Trending hobbies during the pandemic

This past year has seen a rising interest in long-lost hobbies due to shelter-in-place, social distancing, and lockdown orders. Google Trends and Polygraph charted the hobbies that saw the biggest spikes each day of the year.

I’m surprised that sourdough or bread-making is on there, but maybe they didn’t fall under the hobby definition they used.

Tags: , , , ,

Exploring your Google search history

Search history can say a lot of about a person, like where they’re going, where they want to be, what they want to learn about, or what they’re trying to make — at some point in their life. Search Record, by Jon Packles, is a way to parse through your history. Download your archive, import it into the locally-run tool, and explore.

I’m more of DuckDuckGo person, so I can speak to the specificity of the tool, but it looks insightful. At the least, I’d want to download my search archive and play around with it.

Tags: , ,

Blob Opera is a machine learning model you can make music with

David Li, in collaboration with Google Arts and Culture, made a fun experiment to play with:

We developed a machine learning model trained on the voices of four opera singers in order to create an engaging experiment for everyone, regardless of musical skills. Tenor, Christian Joel, bass Frederick Tong, mezzo‑soprano Joanna Gamble and soprano Olivia Doutney recorded 16 hours of singing. In the experiment you don’t hear their voices, but the machine learning model’s understanding of what opera singing sounds like, based on what it learnt from them.

So smooth. So blobby.

Tags: , , , ,

Google search trends dataset for Covid-19 symptoms

Google released a search trends dataset earlier this month. Using this dataset, Adam Pearce made an explorer to compare search volume over time:

The COVID-19 Search Trends symptoms dataset shows aggregated, anonymized trends in Google searches for more than 400 health symptoms, signs, and conditions, such as cough, fever and difficulty breathing. The dataset provides a time series for each region showing the relative volume of searches for each symptom.

Even if you’re not keen on analyzing Covid-19 data, this is likely a good time series source to at least bookmark for later.

Tags: , , ,