Hello,
I have the following server available for transfer:
CPU: Intel Core i5-13500
RAM: 64 GB DDR4
NVMe: 1.536 TB (3×512 GB)
Monthly fee: €54/mo (ex VAT/IP)
Looking for €80 transfer fee.
Seriously I have led $10m+ software development projects for the Department of Defence with some pretty significant security procedures but the ridiculously over the top security hardening Sol tries to push on even small projects is still much more extreme.
I'm pretty sure OpenAI themselves aren't implementing this level of security hardening on their own work.
I'm getting tired of constantly telling it to stop overengineering.
Honestly the Luna and Terra price cuts have just put Terra in an even more awkward position, hasn't closed the price/performance gap enough compared to Sol while it's widened to Luna.
As cheap as Luna is, it's really only useful for small tasks since you have to use xhigh/max and burn through a lot of context.
Would be nice if Terra was a bit cheaper, I think 25% is enough, it won't compare well to Luna on benchmarks for price/performance still but when you need to do more than something Luna can handle it would be useful and compare pretty well against Sol on price/performance.
At this point I still think Sol medium is the best default with Sol high for some harder tasks and Luna Max for some small tasks.
I have this server available for transfer:
CPU: AMD Ryzen 7 7700
RAM: 64GB DDR5
NVMe: 5TB (1x2TB, 3x1TB)
Monthly fee: €74/mo (ex VAT/IP)
Transfer fee: €100
Can give invoice if required.
I have this server I no longer need in HEL if anyone is interested
CPU: AMD Ryzen 7 7700
RAM: 64GB DDR5
NVMe: 6TB (2x2TB, 2x1TB)
€75/mo (ex VAT/IP)
DM me an offer
I'm pretty confident Popa is going to hit us with a big rotation with his obsession with fitness, plus players with a yellow card potentially being suspended for the first knockout match if they pick up another today.
GK: Ryan
LWB: Behich
CB: Degenek
CB: Herrington
CB: Geria
RWB: Trewin
CM: Irvine
CM: Okon
LF: Irankunda
RF: Volpato
CF: Yengi
Personally I think our best 11 for the match is:
GK: Beach
LWB: Bos
CB: Herrington
CB: Souttar
CB: Circati
RWB: Metcalfe
CM: O'Neil
CM: Okon
LF: Irankunda
RF: Volpato
CF: Toure
I've been looking on the hetzner forum marketplace and most listings want you to DM the seller to get the transfer price so it's hard to know what's normal, eg if a server is saving me $30/mo compared to next best option elsewhere, how many months of savings is typical to pay for a transfer?
I don't know if this is just me since no one else is complaining, maybe everyone is playing with Fable. But codex seems genuinely broken and unusable for once (I've never felt that way before). Constant reconnecting, repeating messages multiple times, extremely slow, etc.
​
​
Is it just me or is codex extremely slow the last day or so? I'm struggling to get through my usage because it's so slow (never an issue normally)
29 players left, 3 get dropped - so one of each pair: Gauci/Beach, Geria/Degenek, Borello/Velupillay.
Ignore the positions below a little bit, a few players are versatile across midfield/forward lines.
Robertson being dropped is the big surprise for me, otherwise nothing has really changed from what I would have thought a few weeks ago outside of the McGree injury.
I know we had a reset a few days ago but that was only for Pro and had no announcement so I assumed it wasn't the 5m users reset, but 5m users should have been hot by now so maybe it was?
This can probably be resolved with a prompting/workflow change but I've noticed 5.5 compared to 5.2 and 5.4 doesnt really check the codebase deeply, it does a shallow scan and makes assumptions based on it rather than digging through the codebase in depth. Eg if I ask how does feature X work, 5.2 and 5.4 would dig through the codebase and explain correctly, sometimes they might miss something buried in the code but for the most part they went into depth and got it right, whereas 5.5 seems to just look at the initial code path and then say it 'probably' does Y without checking its correct. This is also evident in its code when it doesnt follow established repo patterns for example.
It probably just needs more prompting and more formalised workflows like in the pre-5.2 era, but a noticeable difference from 5.2/5.4 where you could be lazy with prompting and the model would go through the code and figure things out itself, and kind of annoying that you have to learn these differences and change your workflows when a new model comes out.
This probably explains some of the higher speed and also the surprisingly low usage (given its meant to be much more expensive) of 5.5 as well.
Anyone else finding codex with 5.5 slow currently? It's been nice and fast since release, first time seeing it slow like this.
Since 5.5 Pro was launched I'm constantly getting thinking failed on tasks after it runs for 100+ mins seemingly doing nothing. Usually after trying 3-4 times it will work finally and give a response. Anyone else running into this?
