Petal Search

Petal Search, an English search-engine by China’s Huawei and apparently with its own index. After cleaning all the unwanted cruft off the front-page with uBlock, it can be made nicely minimalist…

Images seemed the most useful. But it turns out the ‘HD’ filter is puny, regarding a mere 900 x 1200 as ‘high’. Still it’s possibly useful as a third-opinion on images, as it gives very different results than DuckDuckGo Images or Google Images.

News feels like Bing, but without the extreme timeliness and with a whole lot of local British news seeming to be filtered through a cheesy relayer called rather than going directly. At a guess that may be to comply with Chinese government requirements?

The main search is sprinkled with ad-cards, but these are easily removed with uBlock. Definitely not as good as Google for the first page. I suspect the problem is in running weaker semantics on the query rather than in the index.

One thing it is is fast. Very fast. They’ll be using their own ‘special’ routers, no doubt.

More UnSplash

A few years back I made three curated picks from the Unsplash CC0 Creative Commons collection, and posted these here…

* Libraries and archives theme.

* Creative Industries I theme.

* Creative Industries II theme.

The CC0 status was later changed. For instance, Alex Knight uploaded his Robot picture under CC0 Creative Commons, prior to the end of 2017. But Unsplash now has its own licence, under which it is claimed this formerly CC0 Creative Commons picture now sits.

The new licence is not that bad actually, on glancing through it… it seems to just prevents the big stock companies from ingesting en masse and re-selling. But now comes the news that the evil stock agency Getty Images is about to “acquire” UnSplash outright anyway.

So here’s another pick, and under the still relatively permissive non-Getty licence, before the purchase goes through and any changes start happening.

What follows is ‘Creative Industries III’. As before, images reduced a little to make them more wieldy, and a few spamming brands (Apple, Nike etc) were airbrushed away. Photographer names are in the file-name, and should be credited if used in print etc. A few of the older pictures (see my collections above) don’t pop up for me in search, where you might have expected to see them again, and may have since been deleted or moved. Useable “draw-on-the-screen” images are very rare, but I found two good ones.

Creative Industries III:

Daylight code writing (useful, as most such pictures are on black):

All-night code writer:

Guitar maker:

Leather crafts maker:

Glass crafts maker:

Pattern knitter:

Children’s book illustrator:

Graphic illustrator:

Special-interest magazine editor:

General designer using a Cintiq or XP-Pen:

Table-top RPG game designer / miniatures painter:

Pinball table designer:

Children’s party clowns:

Branch librarian / local documentary-maker:

Field researcher (a bit spammy, re: the brand, but the closest I could get which says ‘field research’):

Local history writer:

Philosophers / old books conservator:

Sports vehicle designer:

Vehicle livery colourist:

Hair stylist:

Local TV studio, junior camera operator:

Local TV broadcast station desk-jockey:

Live theatrical event desk-jockey:

Podcast interviewer/presenter:

Synth/trance musician / YouTube celeb:

Young stage drummer:

Young comics reader:

Children’s creative dress-up:

Community dance:


Google’s Live Caption, now on desktop PCs

Isn’t the Internet wonderful. Just this morning I was searching and wondering why is there no audio “automatic transcription” software for desktop PCs? This evening… Google’s Live Caption feature is now available on the desktop PC, via the Chrome browser. For free, and running locally and offline and without a Google login.

To enable real-time live subtitles (aka ‘closed captions’ or ‘live captioning’) as your audio or video plays back, first get the latest Chrome then go…




> arrow icon

Live Caption

…and turn it on in Chrome. At this point a set of speech-definition files will be downloaded, to enable the real-time detection of what’s being said. While you’re waiting, set up the preferences for fonts and colours etc.

Those used to AI sets of 1Gb or more will find the Live Caption’s are downloaded in a few minutes, even on a slow connection. Other than the initial download of the definition files the services work locally on the PC and without a Cloud connection. So far as I’m aware this is the first time such a free service is available without a Cloud-upload being needed, still less in real-time.

For this reason I would expect to see third-party UserScripts relatively soon, to enable the transcription to be easily captured into an editable text file as it plays. The playback / transcription continues to run, even when Chrome is not the focus of what you’re doing on the PC, which should help with scripted capture. Obviously if you want the whole thing you would have to let it play back first, to get a full transcription.

Can a recorded .MP3 be loaded and work? As well as a live stream? Yes, it works very well. A podcast with a 90 year-old guy on a smartphone, and kind of ok-ish voice quality… it handled that well. In real-time.

As you watch it, it occasionally goes back and auto-corrects and seems to be doing this based on word context. So I’m guessing it’s not just speech-to-text, but also text-to-text context tweaking. But it can’t work miracles: “gorilla campaign” rather than “guerrilla campaign” etc. And swearing does get f****** bleeped out with asterisks. It can’t detect different speakers. You can’t copy-paste. Still, it’s going to be very useful, especially if you just want a few paragraphs for a quote. Until we get a capture script, you can do things like screen grab with Microsoft OneNote, which handles small fonts fine and can make text from a screengrab very easily.

Incidentally, if you want to edit your audio files first, the venerable freeware Audacity 3.0 is now available.

Block the Google Doodle from search-results pages

Google is trying to creep its Google Doodle visual-messaging, often unwanted and usually distracting, into the top-left corner of the actual search-results page. Block it in the Web browser with the UBlock Origin addon, and this simple filter command pasted into your custom ‘My Filters list’…

If your Web browser insists on adding http:// to the www. bit of the above address (and thus making it a clickable link) then remove this from the copied line. It should look like this…

There’s also a No Google Doodle UserScript, to remove the doodle from the main landing page.

One-click to remove a verbose site from Google

One-click to remove a verbose site from Google Search results, a new UserScript. Preset for Wikipedia, but the URL can be easily changed to be any verbose website. It should ideally be a website that you usually regularly want to remove from search results, but sometimes want to keep. The script is thus more flexible than a regular list-based site blocker.

It works by re-running the current search, but only an instant after some regex has cunningly inserted the command    into the URL.

Also, yes, I’m aware that my ‘add JURN as a link to Google Search’ UserScript has stopped working. Google has re-labelled the divs on the text links just below the search box. A similar script that allows the current search to be passed to Scholar has also stopped working, as have several similar menu scripts. I’m waiting for one of these scripts to update, and thus to show me how it needs to be fixed.

Bite on this…

“Presumed predatory journals are abundant in oral health”, Journal of Evidence Based Dental Practice, March 2021.

From the 580 potentially eligible journals, 431 dental journals were included [and] 226 PP (52%) [i.e. were PP, “presumed predatory”]

Mention or reporting to be indexed in the Directory of Open Access Journals (DOAJ) database and a journal website with distorted or blurry images were the most influential variables for accurate classification into a predatory category or not.

Slightly badly phrased, on that last quote. The authors seem to want to state DOAJ listing = OK (if the link works and the journal’s DOAJ page can be found), while the inability to resize a logo picture correctly = dodgy. But their abstract’s sentence implies the reverse.

But yes, I’d agree that there does seem to be a curious and innate inability among makers of predatory journals to resize pictures correctly, though my eyes are perhaps better at spotting the subtleties of such things than many. Artists and designers tend to forget how dull most eyes are, in terms of not being able to see such details at a glance. Actually, I guess you might even bypass that problem — by training an AI to search for predatory journals through looking for logos and other images with incorrect proportions (squashed, stretched), or prestigious logos that are blurry when they should be crisp, or which have subtly colour-shifted. Bad use of naff fonts is also a tell-tale, I’d suggest, though that would be more difficult to train an AI for. But even just trawling for the DOAJ logo and a couple of other common logos would be an interesting experiment.


Metaforecast is a new search tool that aggregates various forecasts about the future, as mooted and evaluated on forecasting platforms…

we track ~2100 active forecasting questions, ~1200 (~55%) of which are on Metaculus. There are also 17,000 public models from Guesstimate.

As such Metaforecast can’t do what would be really useful. To track media pundits and university press releases, pinpoint and summarise their various claims about the future, and then over time tell us who is the New Paul Erlich and who the New Arthur C. Clarke.