Sheila Brennan (@Doublegee440) header
Sheila Brennan (@Doublegee440) profile picture

If you think the current lot in Parliament are doing a good job, you’re dreaming. Anti-red tape, pro-common sense

Adelaide, South Australia
Joined December 2021
Photo by Sheila Brennan (@Doublegee440)Photo by Sheila Brennan (@Doublegee440)Photo by Sheila Brennan (@Doublegee440)Photo by Sheila Brennan (@Doublegee440)Photo by Sheila Brennan (@Doublegee440)
Sheila Brennan retweeted
I took a deep dive into the newly released Flint model and compared it to Qwen3.5-35B-A3B. The surprise? Their overarching architectures are strikingly similar, yet they make fundamentally different design bets under the hood. Here are 4 interesting architectural observations 👇 1. Layer Layout: The Repeating Pattern 🧱 Flint uses 49 layers, starting with 2 Full Attention layers, followed by a repeating [3 Sliding Window + 1 Full Attention] block.This strongly echoes Qwen3.5-9B’s [3 Gated DeltaNet + 1 Full Attention] rhythm. 💡 Hypothesis: Early Full Attention establishes global context before transitioning into highly efficient local processing. 2. The Skeleton: No Reinventing the Wheel 🦴 Both models share the exact same high-level Transformer block structure (RMSNorm → Attention → Add → RMSNorm → MoE → Add). They both use a 2048-dim hidden state and 256 routed experts.The real innovation isn't in the skeleton—it’s inside the Attention and MoE modules. 3. Attention: Two Different Bets on Memory 🧠 This is where the models diverge significantly: 🔹 Flint (Sliding Window): Directly accesses the most recent 512 tokens + periodic global retrieval. 🔹 Qwen3.5 (Gated DeltaNet): Maintains a continuously updated, compressed recurrent memory. 👉 TL;DR: Flint favors precise local recall. Qwen3.5 favors evolving, compressed long-range memory. 4. MoE: Routing and Compute Allocation ⚡ Both use 256 routed experts, but their allocation strategies differ: 🔹 Flint: Top-6 activated. Features a larger, ungated shared expert (1024 dim). 🔹 Qwen3.5: Top-8 activated. Features a smaller, dynamically gated shared expert (512 dim). 👉 TL;DR: Flint leans heavier on a robust shared knowledge path, while Qwen leans into dynamic, specialized experts. 🎯 My Biggest Takeaway: Flint and Qwen3.5 share a similar blueprint. Despite its impressive coding benchmarks, Flint has yet to report several important general-purpose evaluations, including MMLU-Pro, GPQA, HLE, and long-context benchmarks. Without them, it's hard to conclude that Flint's architectural choices are inherently superior to Qwen3.5's. As Recursive’s debut model, Flint’s architectural design feels surprisingly conservative. Instead of wild innovations, they played it safe with a masterful recombination of proven components. Architecture breakdown below 👇
🤖 Made with AI
62
16
325
37,943
Sheila Brennan retweeted
Kai Cenat explains why he will not be playing GTA 6 unless he's paid by Rockstar , says that if he doesn't play his fans won't play either so they better offer him a contract 😮
14
8
167
58,705
Sheila Brennan retweeted
For an AI car-ad concept, I’d send a Tesla up a wall, through a flip, and into a clean product reveal. Quite an entrance. 🚘⚡ The CapCut plugin is coming soon to ChatGPT desktop! CapCut World brought leading creators and OpenAI together to explore new video workflows. #CapCutWorld #CapCutDidThat
19
12
108
70,459
Sheila Brennan retweeted
Jay Cinco couldn't believe it when someone posing as a janitor served him a lawsuit from his dad who he never met saying he was entitled to 60% of his earnings since he's his son 😮
11
10
179
77,182
Sheila Brennan retweeted
Jay Cinco couldn't believe it when this employee's wife was in Jay's Twitch chat talking to him and he seemed heartbroken to find his wife talking to another man 😢
15
19
230
70,957
Sheila Brennan retweeted
Marlon reveals he has been hired by the LA chargers NFL team to be their mascot , says that they only pay $275,000 a year but that it's purely for the clout 😮
18
5
211
52,875
Sheila Brennan retweeted
50
11
46
689
Sheila Brennan retweeted
Here's your Halloween-season Black Friday reminder🎃CapCut Desktop is offering up to 60% OFF Pro and 40% OFF Ultra through October 30. Desktop only! Picture a Halloween product teaser with creepy close-ups, a sudden reveal, and several quick cuts.
23
15
244
163,135
Sheila Brennan retweeted
Who knew an existing video could be a starting point for so many creative ideas? 😂 Gave this one a new twist with Pollo Recast. Made with @itsPolloAI
🤝 Paid partnership
11
12
98
57,064
Sheila Brennan retweeted
14 campaigns. 30 operational problems. And I had no idea how much time we were wasting keeping track of everything. Payment updates buried in conversations. Missing information scattered across different messages. Deadlines that required constant manual follow-ups. So I decided to give our new AI employee one responsibility: Find out what’s slowing down our campaigns and organize a way forward. I didn’t explain every step. I wanted to see how he’d approach the problem. Here’s what Viktor delivered: → 14 campaigns reviewed.
 → 30 operational issues identified.
 → 6 urgent problems prioritized.
 → $3,050 to $4,050 in completed work without confirmed payment.
 → A complete campaign dashboard.
 → A 9-sheet Excel tracker. He also prepared payment follow-ups and organized the next actions requiring approval. But the biggest result wasn’t the dashboard. It was the time we got back. By centralizing campaign tracking, reporting, and payment follow-ups, Viktor has already should save us over 10 hours of manual work per week. More than 10 hours we can now spend developing new partnerships and growing the business. And there’s something I really appreciate. He didn’t claim the $3,050 to $4,050 had been recovered. He flagged the missing payment confirmations and gave us a clear plan to investigate them. Every important action still requires human approval. This changed how I think about working with AI. The interesting part isn’t how many answers you get. It’s how much responsibility you can give an AI employee and what he actually delivers. Want an AI employee working toward your team’s goals? Hire @viktor_com and start free. #AIemployee Paid Partnership
🤝 Paid partnership
20
26
1
156
90,743
Sheila Brennan retweeted
Gotham wasn’t ready for this Halloween glow-up! 🎃🦇🔥 AI is taking creativity to a whole new level! 👻
🤝 Paid partnership
Gotham wasn’t ready for her 🎃🦇 I’ll drop the template in the replies—and give you free credits to create your own. 👻⬇️ #Wizstar #AIVideo #Batman #halloween
30
50
357
27,117
Sheila Brennan retweeted
Ray couldn't believe it when he got hit by a drunk driver live on Twitch damaging his custom 1 of 1 $3,000,000 Tesla gifted to him by Kai Cenat , says he's going to have to spend an enitre week's paycheck to fix it 😮
17
4
230
88,630
I enjoy coding, I enjoy game design, I enjoy modeling, I enjoy putting it all together and see it come to life. I might be slow, and I might not be the best dev in existence, but I enjoy it. And I want to learn and improve. If I fall behind, so be it. This is not a race.
87
707
25
4,463
43,549
Sheila Brennan retweeted
We are glad to see he's on the winning team , let's go Roobet 🥳 🔥
🤝 Paid partnership
Welcome Home @Cabrzytv 🦘
9
4
45
8,393
Sheila Brennan retweeted
Replying to @gamdom_beekay
$10k for just interacting? Beekay you're the undisputed goat 🐐👑 #Gamdom to #1 🆔 Bluemizu
12
11
11
233
Sheila Brennan retweeted
Tylil couldn't believe it as he was escorted by security from this concert's backstage as soon as they found out about his allegations 😮
15
8
285
73,078
Sheila Brennan retweeted
The gameplay looks intense, every move feels like it matters. Hellraiser: Revival definitely keeps you on your toes. #hellraiserrevival #seesuchsights
3
35
6,815
Sheila Brennan retweeted
recast the bit where you put someone from your life into a scene they have no business being in, and it looks completely real
Why reshoot when you can recast? Same motion. Same timing. New character. Upload an existing video and a reference image of your new character to recast it in ElevenCreative.
5
9
77
17,212
Sheila Brennan retweeted
Ray reveals he has been losing viewership after the allegations, says that he only makes $50,000 a month and can't even afford to pay his mom's rent or bills anymore 😮
15
8
253
77,254