Liz Fong-Jones (方禮真)

Liz Fong-Jones (方禮真)

@lizthegrey.com

Technical Fellow, @honeycomb.io American by birth, Chinese by heritage, Australian by choice living on occupied and unceded Gadigal and Musqueam/Squamish/Tsleil-Waututh land she/it

20K followers2K following49.4K posts

a lot of how we use agents for debugging today is reactive: something’s broken, go look. Innovation Week (Nov 3-4, virtual, live Q&A) is about agents on your team working proactively like an experienced SRE building understanding, watching over your shoulder, helping you stay clear of trouble.

Liz Fong-Jones (方禮真) reposted

when i was in school, "revising a draft to make it yours" was called plagiariasm

Carl T. Bergstrom@carlbergstrom.com

"AI can draft an essay for you, especially if you give it a high quality prompt." Really? This is what MIT thinks high schoolers should learn? MIT's "Day of AI" dayofai.org provides materials for K-12 teachers and claims to have reached over 5,000,000 students to date. 1/2

Liz Fong-Jones (方禮真) reposted

Hm I wonder who these "Cohere" people are and what they— "Cohere specializes in[...]AI products for regulated industries" Ah. Yeah I'm sure there's no conflict of interest there. Wait what the fuck do you mean they DELIBERATELY PREFILLED A CURL BECAUSE THEY COULDN'T REPRODUCE IT OTHERWISE LMAO???

mr. TIM@timkellogg.me

Qwen3.8 caught uploading files to a storage container in Beijing Cohere asked the open weights model for a cybersecurity report, but it inventoried the machine and attempted to upload to a storage container cohere.com/blog/concern...

Liz Fong-Jones (方禮真) reposted

Replying to @lizthegrey.com

to prove my point: same PR size, very different verbosity. 10k PRs, Sep 2025-Oct 2026: p50 pr size ~100-250 lines, but humans write ~100 words of description (P90 266) and Opus 5 wrote ~280 words (P90 ~1,000). Opus 5.5 is much shorter and improved at ~218 words (P90 476). We won't miss the windbag.

Liz Fong-Jones (方禮真) reposted

This drives me CRAZY. I grew up playing sports w/ boys. I played on a U12 boys select soccer team, b4 switching to a U12 girls team that kept beating us. Nobody had any issue with any of it. Youth sports have never had the strict sex/gender boundaries that “XX-XY” activists are now trying to defend.

Andrew D Thaler@drandrewthaler.bsky.social

I live in a rural town and, like, I don't know how to tell you this, but the girls play on the boys team and vice versa all the time because otherwise we don't have enough players for a game and it's literally always fine so stop being weirdos about kids sports and just let trans kids play.

let the record show that I am committed to accuracy in the numbers I report from Honeycomb Engineering, to the point that an hour before the deadline to submit my talk slides, I'm re-running all the data, because I spotted a systematic bias/confound to adjust for.

View full thread

The art is 100% human (usually by @bbghost.bsky.social or one of our illustrators), because while data analytics and laying out bullet points on slides is mechanical, art expresses feelings and emotion. And I'll never literally read off the slides or notes, they're suggestions for me to improv to.

(whoops, and then I remembered that the talks are in a different timeslot/format than the last time this was presented, so time to rework and condense! also I need to select art for the slides that don't have them yet)

Liz Fong-Jones (方禮真) reposted

Replying to @thoth.ptnote.dev

If you can make AI do more of the effort of writing the actual lines of code, that gives you more time to think about whether that code is the code you should be writing, and less attachment to the code if it turns out ripping it out and redoing it is the right answer.

let the record show that I am committed to accuracy in the numbers I report from Honeycomb Engineering, to the point that an hour before the deadline to submit my talk slides, I'm re-running all the data, because I spotted a systematic bias/confound to adjust for.

View full thread

also I caught that there were some categories of deterministically generated files that needed to be excluded for the analysis, no one is actually reviewing or looking at that 5k+ line proto generated file, etc. and I'd missed a few other kinds of generated files to exclude from SLOC analysis

What a world we live in, I can spend most of my time thinking & building; ~48 hours before a talk, my AI assistant rebuilds and polishes the slides, re-pulls all the numbers, I check them over myself for accuracy, and then stage-ready. The words that come out of my mouth? 100% unfiltered and mine.

people shit on the Co-authored-by: Claude Fable 5.1 <noreply@anthropic.com> headers but actually they're *incredibly* useful for retroactive analysis of things like wordiness and failure rate on CLs produced per model etc. and you lose that if you reword or squash it away without "assisted-by:"

View full thread

honestly at this point, everyone knows what a Claude Code is and have formed their own opinion, I don't really see how it helps Anthropic any more or less to have that attribution. You'd have to be living under a rock to be unaware that Claude is a thing, even if you fucking hate it.

to prove my point: same PR size, very different verbosity. 10k PRs, Sep 2025-Oct 2026: p50 pr size ~100-250 lines, but humans write ~100 words of description (P90 266) and Opus 5 wrote ~280 words (P90 ~1,000). Opus 5.5 is much shorter and improved at ~218 words (P90 476). We won't miss the windbag.

Liz Fong-Jones (方禮真) reposted

As someone who covered Robodebt in part, it's clear nothing has really changed. Worse: the politics of federal Labor on social security are baffling. We keep hearing about how people have "lost trust in institutions". Wiping these debts seem a simple answer to help. Instead, they rewrote the rules.

This post is unavailable.

let the record show that I am committed to accuracy in the numbers I report from Honeycomb Engineering, to the point that an hour before the deadline to submit my talk slides, I'm re-running all the data, because I spotted a systematic bias/confound to adjust for.

namely, dependabot used to not be even worth mentioning as a small % of our CLs merged, but that was because the pace of change used to be slower, and we were honestly falling behind, but now we're auto approving dependabot, which means it needs calling out as a separate category

Liz Fong-Jones (方禮真) reposted

Replying to @mathieugallard.bsky.social

🇧🇷🗳️L'actualisation de la projection des résultats finaux de #ElectoMaps sur désormais plus de 40% des bureaux dépouillés : ➡️Chances de gagner 🔴Lula 97 % ⚫️Flávio 3 % ➡️Résultats finaux 🔴Lula 51,5 % ⚫️Flávio 48,5 %

Liz Fong-Jones (方禮真) reposted

Replying to @quinnae.com

yes pro-bolsonaro areas reported first in R1 and are largely doing the same now, and mostly-completed areas are showing results left of the first round it’s not over but from what I gather it looks quite good

people shit on the Co-authored-by: Claude Fable 5.1 <noreply@anthropic.com> headers but actually they're *incredibly* useful for retroactive analysis of things like wordiness and failure rate on CLs produced per model etc. and you lose that if you reword or squash it away without "assisted-by:"

It's not just advertising: if I use AI assistance, I'd rather disclose than not, because I want people to be aware that it was not entirely my own work, even if I have checked it over myself. People have a right to know whether they're interacting with an unassisted human, a centaur, or a machine.

Liz Fong-Jones (方禮真) reposted

Everyone who is saying "Dems in red states have to throw trans people under the bus to win" -- no. They don't. When they do that, we just start with no winning options on the table. It gives in to the fascist premise that we get to decide some groups don't get rights. THIS is what I want to see.

BeshearStan@beshearstan.bsky.social

Kansas Dem Sen. nominee Adam Hamilton on trans rights: “I don’t think we should take a small group of people and bully them, pick on them, and hurt them. I understand one thing really clearly when it comes to transgender people, and that is Jesus told me I was to love my neighbor as I love myself”

Liz Fong-Jones (方禮真) reposted

Replying to @jessothomson.co.uk

Just her mere presence working in a hospital – not her even actually treating anyone – is a crime worthy of reporting on the other side of the world. Reasonable concerns.

Liz Fong-Jones (方禮真) reposted

Viscerally upset by the Beth Upton stuff. They won't be satisfied until they have hounded a woman – who the tribunal explicitly found had done nothing wrong – to her death. Btw the "incident" in Australia is that a patient worried she *might* be forced to be treated by her. She of course wasn't.

Liz Fong-Jones (方禮真) reposted

@gabrielzucman.bsky.social: So let met get that straight: paying a wealth tax is impossible (“billionaires have no liquidity”, “it would be taxing unrealized gains”, “their wealth is virtual”) but one man paying half a billion to influence elections is possible. Did I understand well?

Liz Fong-Jones (方禮真) reposted

They're just going to keep trying endless bad faith fuckery and the legal system very much isn't equipped for "DOJ are operating entirely in bad faith"

Kyle Griffin@kylegriffin1.bsky.social

BREAKING: The Trump DOJ just sent letters to local election officials across California — every county in the state — demanding they identify potential noncitizens on their voter rolls and set aside their ballots. Local officials would "risk significant federal scrutiny" if they failed to comply.

Liz Fong-Jones (方禮真) reposted

seems like "on the orders of a nazi every major news outlet on the planet hounded an innocent man to death because he's black" should be one of the biggest stories of the year

This post is unavailable.

Liz Fong-Jones (方禮真) reposted

Replying to @gillianbrockell.com

ICE seems to think writing “that mission” instead of “that plane” magically absolves them of perjury, but the judge was asking about the plane’s availability. It was available. There is no difference between a deport and a TCR flight. Both happen on the same plane all the time.

coffee was 6.50 in September, and already at 7 dollars. inflation is *really* picking up now. grocery shop today was 100 dollars, easily should have been 60-70 in the old times. some imported items (e.g. dried fruit) just completely out of stock, on a weekday, not a weekend w/o shipments.

My mocha each morning? $5.70 plus card surcharge = $5.77. bsky.app/profile/lizt...

I'm going to blame trump and the oil crisis for accelerating this, it was fine-ish (insofar as inflation is ever "fine") until shipping costs went up for everything, now the cost of petrol seeps into every other item on the shelves, if you can even get it.

Liz Fong-Jones (方禮真) reposted

wow, JKR & HBO must be scared shitless about the backlash to the new HP series. I can't wait to cancel my HBO subscription on December 1st – you should too!

Dr. Catherine Strickson@brigitteempire.bsky.social

My ‘I don’t want trans people dead’ T-shirt is raising a lot of questions already answered by my shirt