HNHacker News
TopNewBestAskShowJobs

sgt101

7,940 karma · joined January 24, 2012

nothing to see here, move along, move along.
submissionscomments
sgt101··on AI Search: The Bitter-Er Lesson
Sure, but there are grounded metrics there (the theorem is proved, not proved) that allow feedback. Same for games, almost the same for domains with cheap, approximate evaluators like protein folding (finding the structure is difficult, verifying it quite well is cheap).

For discovery and reasoning??? Not too sure.

sgt101··on AI Search: The Bitter-Er Lesson
yeah - I think that that's waht they mean and I think that there isn't such a metric. I think people will try to do adversarial evaluation but my guess is that it will just tend to the mean prediction.

The other thing is that LLM inference isn't cheap. The trade off between inference costs and training costs seems to be very application specific. I suppose that there are domains where accepting 100x or 1000x inference costs vs 10x training costs makes sense, maybe?

sgt101··on Cost of self hosting Llama-3 8B-Instruct
Running 13b code llama on my m1 macbook pro as I type this...
sgt101··on The regenerative urban garden I: No-till gardening
Well, I was thinking vs. commercial organic.... and basically home garden labour is free!

But, yeah - farmers are very clever people.

sgt101··on The regenerative urban garden I: No-till gardening
Doing everything at once is a big mistake.

Gardens need infrastructure and they need time to evolve into a system that works and is productive.

Start by building compost bins. Use your mower (if you have a rotary one that bags) to mince up clippings from woodie plants (not actual bits of wood as that will kill the mower) and obviously to collect grass clippings. Start composting this down. Obtain a bag of good live compost or muck full of worms (you can see them) and use that to dope and infect your heap, loads of worm eggs will be in the muck and as the compost gets going the worm population will get going too.

Start your veg patch small and simple. If you have the space eight 3m * 1.2 m beds will be a big challenge to look after at first, but will provide more opportunities to grow things than you will be able to take advantage of (effectively). Remember, growing veg isn't just about preparing the soil, throwing seed on and waiting.

You will need to fence and net against pests. Pigeons will eat all brassicas in short order. Rabbits will eat most stuff, mice and rats as well. Good plants to get going with in the UK are: courgettes and other squashes (pumpkins need lots of room but can be a fun hobby plant), beans (broad, french and runners), peas, potatoes (blight is... a blight), celeriac and cellery, lettuce, broccoli (Calabrese), kale, beetroot, chard, carrots, corn, cabbage. You must net cabbage, broccoli and kale. You will also need to feed - organic seaweed feeds are good for this. Most importantly remember plants grow because of light and water and constant protection and pest management.

Once you are going and learn about the plants then get a green house and expand. The green house and cold frames are essential to get seedlings going - this is when you can move from paying more to grow than it costs to buy to saving money. You can also grow veg that only thrives under glass in a cooler climate. But, be warned.. commercial farmers are very clever and very efficient. You will always struggle to compete with them on a value basis. But it is very satisfying and the food often tastes better than shop bought.

sgt101··on Diffusion on syntax trees for program synthesis
hang on, what if...

we're the bad guys?

sgt101··on Java Language Update – a look at where the language is going by Brian Goetz
>OO was about many things but particularity about code reuse. Is that also a bad thing?

No - but OO wasn't as successful at delivering code reuse as it was promised to be, especially polymorphic OO.

>Software Engineering will not progress into real Engineering, until it starts building on the past instead of throwing away past lessons.

SE won't be real Engineering until we start being able to do things like measuring the robustness of a system, or projecting it's maintenance costs. I think we are as far away from this as we've ever been.

sgt101··on Sharing details on a recent incident impacting one of our customers
Super motivating to have off cloud backup strategies...
sgt101··on Taking Risk
The reality is that when you are trying to sell at scale somewhere outside your domestic market (where your laws and rules run) muscle counts. The USA plays relatively nice, but be under no illusion you will find that the game is stacked against you. Your IP in particular is at risk from domestic actors.

An anecdote: I was once part of an exercise to evaluate the potential for an IP case in the USA to be won. One fun part was that there was a jury simulation. The jury were presented with the case and asked to vote. In the first round the jury voted unanimously for us. In the second round the jury were told that the defendant was a US company and we were a foreign company - and they unanimously voted for the defendant.

In other places you will find that you can't play the game - you can't get licenses to operate, your staff can't work there, your payments don't go through, you get a tax case against you, your staff suddenly fail to show up and it goes on and on.

If you are a USA company that happens less everywhere because the USA has power. People in important places owe the USA their lives, and if not then they have had lovely experiences in the USA. If you are UK company then you operate, now, under the reality that the UK hasn't got that power. World markets are open for the UK - but more or less to the extent that the USA decides that they wish to maintain that openness. Often that suits US interests in places where the UK has no interests and does not suit the USA in places that the UK has extensive interests in.

If you wonder why we are doing Aukus and have two big expensive and hard to defend aircraft carriers then the above is why. We are trying to pay the tariff that the USA wants to tolerate us as part of their system.

But it is the USA's system, cut and dried, over and out.

sgt101··on Taking Risk
This is, at it's heart, about markets.

In the USA a startup's market is, essentially, global - China. In the UK the market is, essentially the UK.

In the recent past it might have been argued that the market was the EU - but the reality is that the EU's market for services is fractured. The USA has the capital and power to enforce their companies access to markets everywhere (apart from China), the rest of the world does not.

sgt101··on Taking Risk
"quantum computers, nuclear fusion, self-driving vehicles, space exploration and drug discovery"

The mark of bullshit. No one can make this judgement.

sgt101··on British engineering giant Arup revealed as $25M deepfake scam victim
Just use a trusted channel?

I mean - we have authentication for bank accounts, why wouldn't that be demanded for transactions like this? Without proper authentication of the authorities there's no way that a transaction like this should be put through.

sgt101··on Dragonfly: An optical telescope built from an array of off-the-shelf Canon lens
I am no expert, but the things I would worry about:

- cost to get into geostationary orbit might dominate the value/saving of the cheap instrument, so it might be smart to spend more on that - managing and controlling it might be very challenging - need to get the data down from it - might create difficulties and costs that kill the value - heating and cooling in space might kill it - radiation in space might kill the hardware - acceleration during launch might kill the hardware - the payload needs to be stable during launch or there will be an accident - scientific value might be lower than other missions for similar spend

sgt101··on Biggest CPU for the Bad System
I ran a stressed app some years ago. We only had a wee little backend because our revenue was v.low, but we wanted to do stuff like sleep inside and eat, and so were motivated to cut costs to make profit.

What I did was make a table of all the queries that were being run on my backend, and I ordered them by the number of times that they were called and the cost of calling them (I honestly can't remember the measure I used for that but it was like cputime*memory or similar). I then did two things for the top queries.

1) Optimised them where I could.

2) Looked for where they were being used and tried to stop it.

(2) was very successful.

sgt101··on 100 Years of IBM
Everyone has outsourced anything that isn't nailed down, that's why not to pay attention to ceo's threatening to outsource you. If they can/could they will/would no matter what. The threats just mean that they can't. There is nothing special abut telco's in this regard.

Telco's don't have leverage to compete on anything but price really, they don't have the capital to innovate, they are prevented from vertical integration by regulation, customers will not pay for service... and it's expensive. So, it's bargain basement time. Their investors have piled them with debt and now are looking at them as an annuity business. One big issue is that humans only have very limited bandwidth requirements and so much networking can be done in the data center vs out on the edge now. This means that I don't really see another generation of edge networking coming along and realistically this does reduce telcos to the old utilities that they once were.

Time for renationalisation?

sgt101··on 100 Years of IBM
>I work in a Telco Fortune500

Sorry to hear that friend ;o) I hope that the pay is good/useful and your work environment is positive.

>great argument for why telcos didn't innovate.

I really think it is all down to not having the money. Regulation, legacy costs and costs required to sustain the core business have drained telcos of all cash. I think that this has been especially painful for Europe and is political - for example in the UK there was even a special tax to claw back £600m from BT at one point. Add in the costs of 3G licenses which must have been £30b or more overall then you can see why Europe doesn't have a competitive technology sector built on it's telco heritage. This all happened during the .com boom when Amazon and Google were evolving - imagine what would have happened in Europe if even 30% of that cash had been available to support innovation.

sgt101··on Cubic millimetre of brain mapped at nanoscale resolution
I was really interested to see that a single neuron has 1000's of exciting connections and 1000's of inhibitory connections. I know that this is a gross feature but it's a reminder of just how distant NN models are from the biological reality.
sgt101··on Cubic millimetre of brain mapped at nanoscale resolution
they don't and talk about the difficulties in their paper. I found it refreshing to see the standard of frankness and openness in how they address this. But - it's all pretty compelling and will surely prompt and sustain a lot more research investigating these results and data and also creating more in the future.
sgt101··on Cubic millimetre of brain mapped at nanoscale resolution
The manuscript gives some details in the context of the difficulty of obtaining larger useful samples in the future and the difficulty of understanding if a sample is typical or pathological.
sgt101··on 100 Years of IBM
Well, AT&T was so insignificant that the government broke it into 8, 7 regional phone companies and a core continental provider. The core (AT&T) then sold its tech capability (Lucent). One of the bits (South Western) then ate the core and decided to call itself AT&T once more, now it's one of the biggest (13th) companies in the USA again.

However, the internet meant dumb core, smart edge. This, and regulation meant that telecoms companies have struggled to make themselves relevant to consumers. Any attempt to do so (undermining net neutrality) generally inspires rage in everyone. Technical change has lead to the situation now of the core just being what happens in data centres and between data centres - everything else is just noise. Telcos weren't able to raise the funds to invest in cloud at the rate that Amazon, Google and MS were because they had other investment pulls (3,4,5G, FTTH) and much less free cash - and so they got shut out.

sgt101··on Most of Europe is glowing pink under the aurora
How did this compare to the Carrington event?
sgt101··on Whistleblower Josh Dean of Boeing supplier Spirit AeroSystems has died
There's a really good movie about it too.

https://en.wikipedia.org/wiki/Silkwood

sgt101··on Tour of CLIPS (2022)
Thanks for flagging, I updated the readme.

I may download everything again and see if I can get the test scenarios running with modern Java. I am so out of date with Java though - what's the accepted best version to run things with now that Oracle has made the license so eyepoppingly evil?

sgt101··on The Stripper Index: An unorthodox recession measurement
>Senior consultants that solve million dollars problems don't charge that much per hour.

I do.

sgt101··on New findings point to an Earth-like environment on ancient Mars
>But what puzzles me is why we have never followed up with any further experiments to try and detect life?

simples.

If life is definitively detected on Mars there will never be another mission to determine if life is present on Mars. So what clever scientists have figured out is that they need to do all the science that they want before checking for life.

sgt101··on Tour of CLIPS (2022)
Try this one: https://www.tandfonline.com/doi/pdf/10.1080/088395199117513 and https://citeseerx.ist.psu.edu/document?repid=rep1&type=pdf&d...
sgt101··on Tour of CLIPS (2022)
Here's a java implementation of rete that I worked on about 20 years ago https://github.com/sgt101/zeus/tree/master/src/zeus/rete

It might work!

sgt101··on AI leaderboards are no longer useful. It's time to switch to Pareto curves
I have advocated and used pareto fronts as a model selection method for ML for a long while. It's really useful to construct two tests - hard but important, and run of the mill - and plot model performance against each one and draw a pareto front so you can see which of your models are off the edge. In fact if you were to look at Figure 8.3 in "Managing Machine Learning Projects" then you would see this kind of thing!

But, I'm just an old bot shilling for my product.

sgt101··on Atomic nucleus excited with laser: A breakthrough after decades
Essentially distributed consistency and co-ordination. NTP isn't accurate/consistent enough because of light speed being limited. This matters for applications like network management and large scale control.
sgt101··on The Financial Times and OpenAI strike content licensing deal
The difference is that mega-corps are ripping off content and reselling it. I hope that the FT can sustain itself for a long time using the money from this deal.
← PreviousPage 17 of 34Next →